Hardware acceleration for segmentation of message packets in a universal serial bus peripheral device

[0002] This application claims the benefit, under 35 U.S.C. §119(e)(1), of U.S. Provisional application No. 60/166,891 (TI-29871PS), filed Nov. 22, 1999, and incorporated herein by this reference.

[0003] Not applicable.

[0004] This invention is in the field of data communications, and is more specifically directed to Asynchronous Transfer Mode (ATM) data communications by way of a Universal Serial Bus (USB) peripheral device.

[0005] The present embodiments relate to universal serial bus (“USB”) systems, and are more particularly directed to increasing the rate and flexibility of communications between a USB host and a peripheral.

[0006] USB is a recently-developed technology established by a joint effort of various companies, resulting in an adopted standard set forth in Universal Serial Bus Specification, Revision1.1, Sep. 23, 1998, which is hereby incorporated herein by reference. The USB Specification is directed to a goal of improving the user-friendliness of various aspects of computers and the peripheral devices typically used with such computers, and to this end governs many aspects about USB systems. In a USB system, peripheral devices are coupled to the host personal computer or workstation computer in a tiered-star topology over the USB bus; in this topology, external devices are physically connected to the USB bus by way of a standardized USB cable, rather than by way of specialized serial and parallel ports. The USB bus is mastered by a USB host, resident in the host personal computer or workstation, with the USB peripherals operating as slave devices on that bus.

[0007] The USB technology provides significant advantages to the computer system user, including the ability to connect up to 127 peripheral devices, in a “daisy-chain” tiered-star topology, to a single USB port on the host computer. The USB technology also permits the user to connect and disconnect USB peripheral devices to or from the USB system without requiring system power-down, and generally with little or no configuration input required from the user. This capability provides considerable flexibility and possible cost reduction in comparison to many contemporary systems, particularly those which can only support one peripheral device per port. USB systems also can easily integrate various functions such as data, voice, and video, into the system through a single serial-data transfer protocol, without requiring add-on cards and the availability of their associated mainboard slots. Additionally, the master-slave arrangement permits the relatively high processing capacity of the host computer to perform and manage much of the data processing required for the peripheral function.

[0008] By way of further background, various techniques have been developed in the field of digital communications for routing messages among the nodes of a network. One such approach is referred to as packet-based data communications, in which certain network nodes operate as concentrators to receive portions of messages, referred to as packets, from the sending units. These packets may be stored at the concentrator, and are then routed to a destination concentrator to which the receiving unit indicated by the packet address is coupled. The size of the packet refers to the maximum upper limit of information that can be communicated between concentrators (i.e., between the store and forward nodes), and is typically a portion of a message or file. Each packet includes header information relating to the source and destination network addresses, permitting proper routing of the message packet. Packet switching with short length packets ensures that routing paths are not unduly dominated by long individual messages, and thus reduces transmission delay in the store-and-forward nodes. Packet-based data communications technology has enabled communications to be carried out at high data rates, up to and exceeding hundreds of megabits per second.

[0009] A well-known fast packet switching protocol, which combines the efficiency of packet switching with the predictability of circuit switching, is Asynchronous Transfer Mode (generally referred to as “ATM”). According to ATM protocols, message packets are subdivided into cells of fixed length and organization, regardless of message length or data type (i.e., voice, data, or video). Each ATM cell is composed of fifty-three bytes, five of which are dedicated to the header and the remaining forty-eight of which serve as the payload. According to this protocol, ATM packets are made up of a number of fixed-length ATM cells, currently limited to a maximum of sixty-four cells. The fixed size of the ATM cells enables packet switching to be implemented in hardware, as opposed to software, resulting in transmission speeds in the gigabits-per-second range. In addition, the switching of cells rather than packets permits scalable user access to the network, from a few Mbps to several Gbps, as appropriate to the application.

[0010] The ATM technology is particularly well suited for communications among computers over the worldwide and public medium commonly referred to as the Internet, because of the flexibility and recoverability provided by this packet-based approach. A relatively recent technology by way of which remote, home, or small office workstations can now connect to the Internet is referred to in the art as digital subscriber line (“DSL”). DSL refers generically to a public network technology that delivers relatively high bandwidth, far greater than current voice modem data rates, over conventional telephone company copper wiring at limited distances. As such, DSL modulator/demodulators (“modems”) are now available for implementation with workstations and personal computers for ATM communications to and from the Internet, with data rates provided by DSL modems ranging from on the order of 500 Kbps to 18 Mbps or higher, according to conventional techniques.

[0011] Given the benefits of USB peripherals discussed above, USB-based DSL modems are now being introduced into the marketplace. Conventional USB-based DSL modems rely upon the host workstation or personal computer to perform such operations as segmentation of message data to be transmitted into ATM cells, and also the reassembly of received ATM cells into the message. This reliance upon the host processor is typical for USB-based peripherals, as noted above.

[0012] It is therefore an object of the present invention to provide a USB-based modem that more efficiently utilizes the USB bus for the communication of data than in conventional USB modems.

[0013] It is a further object of the present invention to provide such a modem in which a portion of the segmentation of ATM packets into ATM cells into messages at the modem rather than at the host system, thus reducing the amount of ATM overhead that is transmitted over the USB bus.

[0014] It is a further object of the present invention to reduce the computational burden upon the host system in performing ATM communications.

[0015] Other objects and advantages of the present invention will be apparent to those of ordinary skill in the art having reference to the following specification together with its drawings.

[0016] The present invention may be implemented into a USB peripheral device, in which ATM segmentation logic is embodied within the USB interface. ATM packets and messages that are generated by the USB host, such as a personal computer or workstation, are forwarded to USB endpoints in shared memory of the USB peripheral. The header portion of each packet, including the virtual connection, packet length, and packet type, is read by the ATM segmentation acceleration logic and stored in a register. Transmission of each ATM cell then begins with the transmission of a four-byte header including this information to a digital signal processor (DSP) for transmission over the communications facility, followed by the transmission of some of payload portion of the packet after retrieval from the shared memory. Upon reaching the boundary of an ATM cell, the header information is then again transmitted, followed by the payload portion. Additionally, cyclic redundancy check calculations may be performed by the ATM segmentation logic to further ease the computational burden of the host system.

[0017] FIG. 1ais an electrical diagram, in block form, of a USB-based system into which the preferred embodiment of the present invention is incorporated.

[0018] FIG. 1bis an electrical diagram, in block form, of the system of FIG. 1a, illustrating the USB interconnections among the various elements therein.

[0019] FIG. 2 is an electrical diagram, in block form, of a USB-based peripheral device according to the preferred embodiment of the present invention.

[0020] FIG. 3 is an electrical diagram, in block form, of an exemplary architecture of the USB-to-DSP (digital signal processor) interface in the device of FIG. 2 according to the preferred embodiment of the present invention.

[0021] FIG. 4 is an electrical diagram, in block form, of the digital signal processor interface and ATM acceleration logic in the USB-to-DSP interface of FIG. 3, according to the preferred embodiment of the invention.

[0022] FIG. 5 is an electrical diagram, in block form, of an ATM receive controller in the logic of FIG. 3, constructed according to the preferred embodiment of the invention.

[0023] FIG. 6 is a flow diagram illustrating the operation of the reassembly logic in the ATM receive controller of FIG. 5 according to the preferred embodiment of the invention.

[0024] FIG. 7 is a flow diagram illustrating the operation of the bus logic in the ATM receive controller of FIG. 5 according to the preferred embodiment of the invention.

[0025] FIG. 8 is a flow diagram illustrating the operation of the reassembly logic in the ATM receive controller of FIG. 5 in processing an OAM packet according to the preferred embodiment of the invention.

[0026] FIG. 9 is a flow diagram illustrating the operation of the reassembly logic in the ATM receive controller of FIG. 5 in processing an ATM cell according to the preferred embodiment of the invention.

[0027] FIG. 10 is an electrical diagram, in block form, of an ATM transmit controller in the logic of FIG. 3, constructed according to the preferred embodiment of the invention.

[0028] FIG. 11 is a memory space diagram illustrating the organization of various ATM packets that may be handled by the ATM transmit controller of the preferred embodiment of the invention.

[0029] FIG. 12 is a flow diagram illustrating the operation of the segmentation logic in the ATM transmit controller of FIG. 10 according to the preferred embodiment of the invention.

[0030] FIG. 13 is a flow diagram illustrating the operation of the segmentation logic in the ATM transmit controller of FIG. 10 in generating and transmitting an ATM cell header according to the preferred embodiment of the invention.

[0031] As will be apparent from the following description, the present invention may be beneficially used in connection with many different alternative system implementations. It is therefore contemplated that those skilled in the art having reference to this description will be readily able to implement the present invention in many alternative realizations, over a wide range of electronic functions and systems. As such, it will be understood that the following description is provided by way of example only.

[0032] FIG. 1aillustrates an exemplary USB system 10 into which the preferred embodiments may be implemented. By way of introduction, system 10 includes aspects known in the USB art and further includes the preferred embodiments. System 10 includes a USB host 12 which, in the present example, is a personal computer (“PC”). USB host 12 includes a motherboard (not separately shown) which communicates with USB software that automatically loads device drivers in a manner that is typically transparent to the user of the PC, where the drivers support the remaining devices external from USB host 12.

[0033] As further detailed later in connection with FIG. 1b, for purposes of USB communications, USB peripheral devices may be directly connected to USB host 12, or may be indirectly coupled to USB host 12 through other USB devices that are directly connected to USB host 12, as shown in FIG. 1. More particularly, in system 10 external modem 14, a monitor 16, and a keyboard 18 are directly connected to USB ports of USB host. In contrast, the remaining devices in system 10 are coupled to USB host 12 by way of one of these directly-connected USB devices. For example, spea 22 are connected to USB host 12 by way of USB connections to monitor 16, which in turn has a USB connection to host 12; it should be noted that, typically, video data to be displayed by monitor 16 will not be communicated thereto by host 12 over a USB connection, but will instead be communicated over a separate standard video connection (not shown in FIG. 1aand b). Similarly, mouse 24 and scanner 26 are coupled to USB host 12 via USB connections to keyboard 18, which is a USB device connected to a USB port of host 12. While each of the illustrated peripheral devices in system 10 is shown as a USB device, it is of course to be understood that, alternatively, only a subset of these devices may be USB devices.

[0034] The operation of system 10 includes numerous aspects known in both the general-purpose computer and USB arts. As to the former, a detailed discussion is not presented in this document because the preferred embodiments are particularly directed to USB aspects; thus, briefly it may be appreciated that each of the devices in system 10 operates to perform the known functionality for such a device and with respect to the PC forming USB host 12, such as data input from keyboard 18, mouse 24, scanner 26, and microphone 22, data communication to and from a remote location by modem 14, and data presentation by monitor 16 and speakers 20_Land 20_R. The USB operation of system 10 is detailed throughout the remainder of this document.

[0035] FIG. 1billustrates an electrical diagram of the USB hierarchy of connections among the devices of FIG. 1a. As shown by the legend toward the right of FIG. 1b, each device in the upper half of FIG. 1bis referred to in the art as a hub and, thus, the hubs in system 12 include USB host 12 (i.e., the PC), keyboard 18, and monitor 16. In a USB system, the system includes a singe USB host (e.g., host 12) that also serves as a hub, where this and any other hub in system 10 is a wiring concentrator for connecting to one or more other USB devices. To facilitate such connections, each hub (other than the USB host) includes at least one upstream port through which it is connected either directly to the USB host or to another hub, and one or more downstream ports to which other USB devices may be connected, all connections being achieved using USB cables. Each USB cable includes four conductors, two for providing power to a USB device if the device does not obtain power via some other source, and two for data communications. The connectors on each end of a USB cable differ from one another so as to ensure that a proper end of the cable is connected in the upstream direction toward USB host 12 while ensuring that the opposing end is connected in the downstream direction away from USB host 12. Looking to the connections of the hubs in system 10, USB host 12 is connected from a port 12₁, via a USB cable C₁, to a port 18₁of keyboard 18, and from a port 12₂, via a USB cable C₂, to a port 16₁of monitor 16. Typically, in modern USB computer systems, monitor 16 serves as a USB hub in parallel with its primary function as a display device, and receives its video display signals from host 12 over a direct video display connection and not over the USB connection. As also shown by the FIG. 1blegend, each device in the lower half of the FIG. is referred to in the art as a function, although such devices are sometimes referred to (perhaps less precisely) as peripherals. A function is a USB device that provides a capability to the host. In the present example, therefore, the functions include microphone 22, speakers 20, modem 14, mouse 24, and scanner 26. In addition, however, a USB device functioning as a hub may also itself serve as a function; for example, keyboard 18 is an example of a device that is both a hub and a USB function. Each of the functions is also connected via a corresponding USB cable to a hub. For example, a cable C₃connects modem 14, via a port 14₁, to USB host 12 via its port 12₃. The remaining cable connections in system 10 will be readily ascertanable by one skilled in the art.

[0036] In its general operation, USB host 12 operates in a master/slave relationship relative to each of the functions, where USB host 12 always serves as the master and each of the functions always serves as a slave. Further in this regard, USB host 12 includes a serial interface engine (“SIE”) (not separately shown) that is typically incorporated into a USB controller also included with the host so that USB host 12 may communicate serial information between itself and the functions. Particularly, the serial data passes along the data conductors in the cables shown, where. typically the communications at the data conductor level are referred to as USB communications along a USB bus. To facilitate its master operations, USB host 12 generally includes three software levels which, from highest to lowest are: (1) a host controller driver which links whatever specific type of device that is chosen as a USB host controller to the remaining USB software structure; (2) USB system software that communicates between the host controller driver and client software; and (3) client software which is associated with a particular function and is often provided by the manufacturer of the function so that USB host 12 may communicate with and receive the functionality of that function. Given its software levels, USB host 12 monitors the network created by the connections of system 10, and detects when a function is added thereto (or removed therefrom). More particularly, upon attachment of a function to the network, USB host 12 as master detects the added function, and its communication speed, in response to a resistance change due to one or more resistors connected inside the function and which thereby change the resistive load when a USB cable thereto. In response, USB host 12 electrically configures a port connection to the newly-added function. Next, USB host 12 interrogates the function in connection with a four-step process, referred to in the art as enumeration, to identify information about the function and to assign a unique address thereto. Also in connection with this process, or thereafter, USB host 12 may configure the function. Finally, USB host 12 loads the appropriate driver to communicate with the function, and thereafter USB communications may proceed according to a USB protocol discussed below.

[0037] The USB protocol divides the time of communication along the USB bus into one millisecond frames. During each frame the bandwidth is shared among all devices connected to the USB system, and each frame is subdivided into one or more packets. The use and length of packets are constrained according to various criteria set forth in the USB Specification. In general, USB host 12, as master, begins each frame by communicating a start of frame (“SOF”) packet. Thereafter, communications during the frame occur according to a token protocol, in which a transaction between host and a function occurs in response to the issuance of a token followed by an order of response. Thus, USB host 12 sends a token packet which includes an address directed to one of the functions, as well as an indication of whether the data to be communicated is a read (i.e., from the addressed function to USB host 12) or a write (i.e., from USB host 12 to the addressed function). The address specifically identifies what is referred to in the USB art as an endpoint (or “device endpoint”), which is a uniquely addressable portion of a USB function that is the source or sink of information in a communication flow between the USB host and the function. The endpoint gets its name from the fact that it is typically a location in a first-in-first-out (“FIFO”) memory space of the function, so for data written to the function it is written to the end, or endpoint, of a write FIFO whereas for data read from the function it is read from the end, or endpoint, of a read FIFO. Returning to the token operation, when the token reaches the addressed function, that function decodes the address and identifies itself as the destination. Next, one or more data packets are communicated along the network, where the destination function acts accordingly (i.e., either receives or transmits the data). Finally, once the data communication is complete, the recipient of the data issues a handshake packet to indicate whether the transmission was successful. This handshake indication may be either a positive acknowledgment (“ACK”) or a negative acknowledgment (“NAK”). Further, in the case of a function as a data recipient, the function may provide a handshake indication of a STALL where either the intended endpoint is halted or a control request is not supported.

[0038] The USB Specification (Universal Serial Bus Specification, Revision1.1, Sep. 23, 1998) requires that USB data packet transfers fall into one of four data categories: (1) control transfers; (2) bulk data transfers; (3) interrupt data transfers; and (4) isochronous data transfers. In the prior art, each of these transfer types is performed to a like kind of endpoint. Further, in the art and as a logical construct, each such communication is referred to as along a pipe to the endpoint. For example, if a host communicates an isochronous data packet to a function, then it more particularly communicates it to an isochronous endpoint in the function and is said to be along a pipe to that endpoint. Similarly, if a host communicates a bulk data packet to a function, then it communicates it along a pipe to a bulk data endpoint in the function. One skilled in the art will appreciate the application of this terminology to the remaining data transfer and corresponding endpoint types. Finally, while not fully detailed herein, the USB Specification places different constraints on different ones of the data transfer types, such as the number of bytes permitted per packet and the number of packet per frame or for a given number of frames. Some of these constraints are discussed later in this document

[0039] Control transfers allow USB host 12 to access different parts of a function, to obtain information about the function, and to change the behavior of the function. More particularly, control transfers support configuration, command, and status type communication flows between client software in USB host 12 and a function corresponding to that software. For example, control data is used by USB host 12 to configure a function when it is first attached to system 10. Further, each USB function is required to implement an IN control pipe, with a corresponding endpoint 0, as a default control pipe which is used by the USB system software to write control information to the function, Each USB device is also required to have an OUT control pipe (and endpoint) to output control information. The default IN control pipe provides host 12 with access to information pertaining to a USB function such as its configuration, status, and control information. Further, the USB Specification defines requests that can be used to manipulate the state of a function, and descriptors are also defined that can be used to contain different information on the device. Finally, a function optionally may provide endpoints for additional control pipes for other implementation needs, such as to accommodate implementation-specific functionality provided via customer software on USB host 12.

[0040] Bulk transfers permit communication of relatively large data groups where the data may be communicated at highly variable times and the transfer may use any available bandwidth. Bulk transfers are unidirectional and, thus, a given transfer may be only from host to function or function to host; thus, if both directions are desired, then a function must have both an IN bulk endpoint and an OUT bulk endpoint or, alternatively, two pipes may be associated with the same endpoint. Examples of bulk transfers include the communication of data to a printer (not shown), or receipt of data collected by scanner 26. Error detection is included in hardware and implements a limited number of retries for bulk data transfers so as to greatly enhance the likelihood of successful data delivery. However, a tradeoff involved with the communication of bulk data is the possibility of latency. Lastly, the amount of bandwidth per USB frame allotted to bulk data may vary depending on other bus demands arising from other data transfers by either the same or a different function.

[0041] Interrupt transfers are relatively small transfers to or from a USB function. Such data may be presented for transfer by a function at any time, but because USB host 12 is a master it cannot be interrupted. Instead, USB host 12 periodically polls each function and, in response to a notification that interrupt data has been posted, USB host 12 retrieves the interrupt information. Interrupt data typically consists of event notification, characters, or coordinates that are organized as one or more bytes. For example, interrupt data may be presented by keyboard 18 or mouse 24 (or some other pointing device).

[0042] Isochronous data is continuous and real-time in creation, communication, and use. Isochronous transfers are unidirectional and, thus, can be only from host to function or function to host; thus, if both directions are desired, then a function must have both an IN isochronous endpoint and an OUT isochronous endpoint (or two pipes associated with the same endpoint). Timing-related information is implied by the steady rate at which isochronous data is received and transferred. Isochronous data must be delivered at the rate received to maintain its timing. Isochronous data also may be sensitive to delivery delays. For isochronous pipes, the bandwidth required is typically based upon the sampling characteristics of the associated function. The latency required is related to the buffering available at each endpoint. A typical example of isochronous data would be real-time video information received by modem 14. Due to its real-time nature, the delivery rate of isochronous data must be maintained or else drop-outs in the data stream will occur. Isochronous communications are not corrected such as by hardware retries, with the benefit being that timely delivery is ensured (assuming no other latency, such as in software) with the drawback being that data communication may be lost. In practice, the bit error rate of USB is predicted to be relatively small so that applications using the types of data being communicated as isochronous data are not appreciably affected in a negative manner. Lastly, USB isochronous data streams are allocated a dedicated portion of USB bandwidth to ensure that data can be delivered at the desired rate.

[0043] FIG. 2 illustrates a block diagram of a function card 28 according to the preferred embodiment. Function card 28 represents an electrical computer-type circuit board in general, and in FIG. 2 the blocks shown are those implemented in the preferred manner of forming modem 14 of FIGS. 1aand 1b; thus, function card 28 is intended to be enclosed within the external housing of modem 14 and connected electrically to the USB bus as known generally in the art. Further, while function card 28 includes various inventive aspects detailed below in the context of modem 14, one skilled in the art should appreciate that various of these aspects may apply to any one or more of the other functions in system 10. Lastly, by way of example and also for further introduction, in the preferred embodiment modem 14 is a hybrid modem serving both voice-band (e.g., V.90) and DSL communications.

[0044] Turning to certain connections of function card 28, the USB bus is shown to the left of FIG. 2, and corresponds to the two data conductors of a USB cable, for example the data conductors of cable C3 where function card 28 corresponds to modem 14. The USB bus is coupled to a USB interface device 30 which, as detailed below, includes various other functional blocks that are formed using one or more integrated circuits. USB interface device 30 is further connected to host port interface bus HPIF which is further connected to a digital signal processor (“DSP”) 32 or some other desirable processing circuit. By way of example, DSP 32 may be one of various types of DSPs commercially available from Texas Instruments Incorporated, such as the TMS320C6201, TMS320C6202, or TMS32JC6205. In the case of DSP devices such as the TMS320C6201 that have an external host port interface (“HPIF”), bus HPIF corresponds to such a host port interface bus, which is in this case a sixteen bit bus. On the other hand, DSP devices such as the TMS320C6202 have an Xbus (extended bus) interface that supports various operational modes, one of which is a thirty-two bit host port interface mode. Accordingly, where DSP 32 is implemented by way of a device having an Xbus interface, bus HPIF corresponds to an Xbus bus operating in the host port interface mode.

[0045] DSP 32 is further connected to two different analog front end (“AFE”) circuits, namely, a V.90 (i.e., voice-band) AFE 34 and an xDSL AFE 36. Each AFE 34 and 36 is connected to a physical connector 38 for connecting to the communications facility (not shown) over which modem communications are carried out.

[0046] According to the preferred embodiment of the invention, in which function card 28 corresponds to a hybrid modem, DSP 32 includes some amount of on-chip memory useful for the storage of communication data that has been received or that is about to be transmitted. In particular, as shown in FIG. 2, a portion of this on-chip memory of DSP 32 corresponds to receive FIFO 33, which is a first-in-first-out buffer within which data received from AFE 34 or 36, as the case may be, are stored prior to forwarding to USB interface device 30 and host 12. In the case of Asynchronous Transfer Mode (ATM) communications, receive FIFO 33 stores incoming ATM cells. Conversely, another portion of the on-chip memory of DSP 32 corresponds to transmit FIFO 35, within which data (e.g., ATM cells) are buffered after receipt from host 12 via USB interface device 30, prior to transmission via the appropriate one of AFEs 34, 36.

[0047] Returning to USB interface device 30 and examining the blocks therein, the USB bus connects within USB interface device 30 to a USB interface module 40. USB interface module 40 is further connected to a bus B. Also connected to bus B is a USB-to-DSP controller 42, which is further connected to bus HPIF. Lastly, USB interface device 30 includes a shared memory 44 connected to bus B, and that is given its name because it is accessible by both USB interface module 40 and USB-to-DSP controller 42 via bus B. Further in this regard, shared memory 44 includes various locations reserved as USB endpoints.

[0048] A brief description of the operation of function card 28 is now presented, with further details presented later in connection with a more detailed examination of certain of the blocks therein. In general, function card 28 interfaces at both the physical and protocol levels with the USB system and, hence, permits communications between function card 28 and USB host 12. USB host 12 communicates data along the USB bus to function card 28, and that information is received by USB interface module 40 and processed according to principles known in the USB art. Further in this respect, USB interface module 40 may write transfers of any of the four above-described types to the endpoints in shared memory 44, where such information is written via bus B. In addition, USB-to-DSP controller 42 also may access the endpoints in shared memory 44 via bus B. Accordingly, given this access, data written to the endpoints from USB interface module 40 may be read by DSP 32 via bus HPIF or, alternatively, data written by DSP 32 to bus HPIF may be transferred by USB-to-DSP controller 42 to the endpoints in shared memory 32. Further in this regard, DSP 32 is programmed and configured to provide the general functionality supported as a USB function which, for the present example, is a modem functionality given that card 28 is associated with modem 14. Moreover, DSP 32 is programmed in this respect to support both voice and xDSL communications, and the two AFEs 34 and 36 provide electrical interfaces to and from DSP 32 further in this regard. DSP 32 also may be programmed to communicate with respect to USB-to-DSP controller 42 in various manners according to the preferred embodiments, as will be discussed below.

[0049] Referring now to FIG. 3, an exemplary architecture of USB interface device 30 in modem 14 of FIG. 2, according to the preferred embodiment of the invention, will now be described. Of course, USB interface device 30 may be constructed according to any one of a number of architectures and arrangements. As such, it is to be understood that the exemplary architecture illustrated in FIG. 3 and described herein is presented by way of example only.

[0050] USB interface device 30 of FIG. 3 includes functions similar to those provided by the TUSB3200 USB peripheral interface devices available from Texas Instruments Incorporated, and includes some common architectural features therewith. In this regard, USB interface device 30 includes microcontroller unit (MCU) 100, which may be a standard 8052 microcontroller core. MCU 100 is in communication with various memory resources over bus B, including program read-only memory (ROM) 102, and random access memory (RAM) banks 104, 106. RAM bank 104 is utilized primarily as code space that may be loaded from USB host 12 over the USB bus, or alternatively from another device over another one of the ports provided in USB interface device 30. As will be described in further detail hereinbelow, USB endpoint buffers reside within synchronous RAM bank 106; in this sense, RAM 106 serves as shared memory 44 as shown in shown in FIG. 2. MCU 100, as well as the other synchronous functions of USB interface device 30, are clocked at the appropriate clock rates by phase-locked loop (PLL) and adaptive clock generator (ACG) 110, which generates various frequencies divided down from a reference clock generated by oscillator 108 according to the frequency set by external crystal 109. PLL and ACG 110 preferably provides clocks suitable for supporting the available USB synchronization modes, including asynchronous, synchronous, and adaptive modes for isochronous endpoints.

[0051] For USB communications with USB host 12, USB interface device 30 includes USB transceiver 112, which preferably supports full speed (12 Mb/sec) data transfers, and includes a differential input receiver, a differential output driver, and two single ended input buffers. USB transceiver 112 is coupled to USB serial interface engine (SIE) 114, which manages the USB packet protocol requirements for data transmitted and received by USB interface device 30 over the USB bus. In general, SIE 114 decodes packets received over the USB bus to validate and identify the packet identifier (PID), and generates the correct PID for packets to be transmitted over the USB bus. Other receive functions performed by SIE 114 include cyclic redundancy check (CRC) verification, and serial-to-parallel conversion; for transmit, SIE generates the CRC value and also effects parallel-to-serial conversion. SIE 114 bidirectionally communicates with USB buffer manager (UBM) 116, which controls reads and writes of data from and to the appropriate USB endpoint buffers in RAM 104, 106. In this regard, UBM 116 decodes the USB function address in received packets to determine whether the packet is in fact addressed to USB interface device 30 itself, as well as decoding the endpoint address contained in the received packet (which may include a polling packet from USB host 12). Suspend and resume logic 117 is also provided for detecting suspend and resume conditions on the USB bus, and for controlling SIE 114 accordingly.

[0052] Other various functions are also provided within USB interface device 30. Inter-IC (I²C) controller 122 is coupled to bus B, and supports communications to and from other integrated circuits over a two-wire serial connection; for example, code RAM 104 may be loaded from such an external integrated circuit over the I²C port, under the control of I²C controller 122. General purpose port logic 124 interfaces bus B to general purpose parallel input/output ports, numbering two in this example. Timers 126 provide one or more timer functions for controlling the operation of USB interface device 30. Reset and interrupt logic 128 monitors various interrupt and reset conditions, to provide interrupt and reset control for MCU 100. Additionally, extra internal memory is provided by asynchronous RAM 130, which is externally accessible, for example to DSP 32 by way of a dedicated RAM interface (and thus permitting reads and writes in a manner independently from and asynchronously with the USB functionality of USB interface device 30).

[0053] In addition to the USE, interface functions described above, which are substantially common with the TUSB3200 USB peripheral interface devices available from Texas Instruments Incorporated and which effectively correspond to USB interface module 40 of FIG. 2, USB interface device 30 according to the preferred embodiment of the invention includes DSP interface and ATM acceleration logic 120. DSP interface and ATM acceleration logic 120 processes data received from USB host 12 over the USB bus for application to DSP 32 by way of VBUS-to-HPIF bridge 118, and conversely processes data received from DSP 32 before transmission to USB host 12 over the USB bus. The construction and operation of DSP interface and ATM acceleration logic 120 will be described in further detail hereinbelow. VBUS-to-HPIF bridge 118 supports reads and writes to on-chip memory of DSP 32 in either a sixteen bit or thirty-two bit mode. Referring back to FIG. 2, DSP interface and ATM acceleration logic 120, in combination with VBUS-to-HPIF bridge 118, implement USB-to-DSP controller 42 within USB interface device 30.

[0054] Referring now to FIG. 4, the construction of DSP interface and ATM acceleration logic 120 according to the preferred embodiment of the present invention will now be described. As shown in FIG. 4, multiple controllers within DSP interface and ATM acceleration logic 120 are coupled to bus B (FIG. 3). According to this embodiment of the invention, in which USB interface device 30 is implemented into modem 14, DSP interface and ATM acceleration logic 120 includes ATM transmit controller 132 and ATM receive controller 134, each of which is coupled between bus B and controller 140, and are utilized to carry out ATM communications processing, including such functions as segmentation and reassembly, respectively. The operation of ATM transmit controller 132 and ATM receive controller 134 will be described in further detail hereinbelow, relative to the preferred embodiment of the invention. Host interface controller 135 is bidirectionally coupled between bus B and controller 140, while code overlay controller 136 unidirectionally communicates data (corresponding to program instructions for DSP 32) from bus B to controller 140. According to the preferred embodiment of the invention, each of controllers 132, 134, 135, 136 includes an interface to MCU 100 (FIG. 3), by way of which controllers 132, 134, 135, 136 are configured to point to the corresponding assigned USB endpoint buffers in shared memory 44 (e.g., in RAM 106 of the implementation of FIG. 3).

[0055] Controller 140 arbitrates access by controllers 132, 134, 135, 136 to bus VBus (which appears as a “virtual” bus to devices external to USB interface device 30), and further permits access to VBUS-to-HPIF bridge 118 and internal registers 138, as slaves on bus VBus. Specifically, in response to one of controllers 132, 134, 135, 136 issuing a request to master bus VBus, controller 140 operates to grant access to bus VBus according to a corresponding bus protocol, after arbitration among competing bus requests. The bank of internal registers 138 also communicate with controller 140, and resides as a slave on bus VBus, for storing configuration information for DSP interface and ATM acceleration logic 120 and its functional modules. As noted above, endpoint buffer information is preferably configured internally to each of controllers 132, 134, 135, 136; the configuration information stored by internal registers 138 includes such other configuration and status information as appropriate for the operation of DSP interface and ATM acceleration logic 120, for example in the manner described hereinbelow.

[0056] VBUS-to-HPIF bridge 118 operates effectively as a bridge between the bus connected to DSP 32 and internal bus VBus which, as noted above, is accessible to controllers 132, 134, 135, 136. For example, if the interface from USB interface device 30 to DSP 32 is by way of its host-port interface (HPIF), such as incorporated into the TMS320C6201digital signal processor (DSP) available from Texas Instruments Incorporated, VBUS-to-HPIF bridge 118 operates effectively as a bridge to that interface, formatting and translating the communicated data signals from those on bus VBus into a format appropriate for the bus protocol of the host-port interface.

[0057] ATM Cell Reassembly

[0058] As noted above relative to FIGS. 2 and 3, modem 14 in system 10 according to the preferred embodiment of the invention provides a USB-based DSL modem, over which ATM packets may be transmitted and received. According to the preferred embodiment of the invention, ATM packets are received by modem 14 and are reassembled in USB interface device 30 to a significant extent, thus relieving host 12 from performing a large portion of the computationally intensive reassembly operations, which are conventionally performed by host software, Specifically, it is contemplated that USB interface device 30 is able to efficiently perform the ATM reassembly operations of VPI/VCI lookup, OAM cell filtering, and CRC check, Additionally, according to the preferred embodiment of the invention, USB interface device 30 is able to receive ATM cells, in their fifty-three byte form with five bytes of header, and to forward only the payload portion of the received cells over the USB bus. The efficiency of USB communication is thus improved not only by not transmitting the ATM headers thereover, but also by concatenating the forty-eight byte payloads of adjacent ATM cells into sixty-four byte USB packets, thus greatly improving the ratio of effective data to protocol overhead communicated over the USB bus. Additionally, because the USB interface device reassembles ATM packets directly into host memory by way of USB endpoints, and then notifies the host of the completion of the reassembly of a packet, no on-chip memory is required for the USB interface device beyond the USB endpoint FIFOs, thus enabling the implementation of the present invention into relatively low cost peripheral devices.

[0059] Referring now to FIG. 5, the construction of ATM receive controller 134 according to the preferred embodiment of the present invention will now be described in detail. Of course, while the preferred implementation of ATM receive controller 134 is illustrated in FIG. 5 by way of example, it is contemplated that variations in the construction and architecture may be utilized by those skilled in the art having reference to this specification within the scope of the present invention.

[0060] As shown in FIG. 5, and as discussed above relative to FIG. 4, ATM receive controller 134 is connected to bus B in USB interface device 30. According to the preferred embodiment of the invention, ATM receive controller 134 is associated with particular USB endpoints resident in shared memory 44. The particular memory addresses of these endpoints are stored in configuration registers 52 of ATM receive controller 134, written thereto by MCU 100. For control purposes, completion endpoint 51₀is a bulk endpoint that is associated with ATM receive controller 134. According to this exemplary embodiment of the invention, ATM receive controller 134 is capable of handling multiple ATM connections simultaneously. In this example, three endpoints 51₂, 51₃, 51₄in shared memory 44 are accessible to ATM receive controller 134, to which ATM payload corresponding to three virtual connections (VCIs) may be written. Filter endpoint 511 is also accessible b ATM receive controller 134, and is configurable to receive Operation and Maintenance (OAM) cells; alternatively, endpoint 51₁may be configured to receive ATM payload for an additional VCI (e.g., VCI₀), in which case ATM receive controller 134 supports four virtual connections. According to this preferred embodiment of the present invention, each of endpoints 51₁, 51₂, 51₃, 51₄is a bulk USB endpoint. Bulk USB implemented as at least two sixty-four byte buffers arranged according to a conventional FIFO scheme to permit substantially continuous access, such as a “Ping-Pong” arrangement of a pair of buffers, or an elastic FIFO if three or more buffers are provided. Alternatively, some or all of endpoints 51 may be isochronous endpoints, if latency is an issue. In any case, endpoints 51 in shared memory 44 are much smaller in capacity than would be required for the storage of a maximum size (64 k byte) ATM packet, as the packet reassembly according to the preferred embodiment of the invention is effectively made into host memory by way of these endpoints 51, rather than into memory within USB interface device 30.

[0061] According to the preferred embodiment of the invention, the operational logic of ATM receive controller 134 is provided by way of three state machines, namely receive state machine 50, reassembly state machine 60, and VBus state machine 70. Each of state machines 50, 60, 70 is preferably realized by way of sequential logic, suitable for performing their respective functions as will be described hereinbelow. In this regard, receive state machine 50 controls the data communication to endpoints 51 over bus B, reassembly state machine 60 controls the performing of various ATM reassembly functions as will be described hereinbelow, and VBus state machine 70 controls the issuing of bus requests from, and receipt of bus grant signals by, receive ATM controller 134, as well as the communication of data from bus VBus. It is contemplated that those skilled in the art will be readily able to implement the appropriate logic for carrying out these functions, from the description of their operation as set forth hereinbelow.

[0062] As noted above, receive state machine 50 controls the data communication to endpoints 51 over bus B, specifically by receiving data bytes from byte buffers 58 via multiplexer 56 and by applying the bytes to bus B in combination with the appropriate memory address for the corresponding endpoint 51 in shared memory 44. In this regard, receive state machine 50 accesses configuration registers 52 that contain the endpoint memory addresses, as noted above. Receive state machine 50 is enabled and otherwise controlled by the state of bits set by MCU 100 in control/status registers 53; receive state machine 50 can also set status flags in registers 53 for response by MCU 100, for example to initiate exception handling by MCU 100. As evident from FIG. 5, each of four byte buffers 58 stores one byte of a thirty-two bit data word, and provides its stored contents to multiplexer 56 and thus to receive state machine 50. Receive state machine 50 is also in communication with current buffer pointer 54 by way of which the appropriate buffer in the FIFO arrangement of the corresponding endpoint in shared memory 44 is accessed, and with buffer byte counter 55 that maintains a byte count for that buffer, as used in carrying out a handshaking protocol with USB buffer management circuitry (e.g., UBM 116 of FIG. 3).

[0063] Reassembly state machine 60, according to the preferred embodiment of the invention, controls various ATM reassembly functions that are performed by receive ATM controller 134 in USB interface device 30; as noted above, these reassembly operations are performed by the USB host, according to software routines, in conventional USB-based systems. These reassembly functions include calculation of CRC checksums (CRC-10 for OAM cells, and CRC-32 for AAL5 packets) as performed by CRC logic 64. Forwarding of received data is controlled by reassembly state machine 60 according to information stored in DMA table 66. DMA table 66 is a table of entries (e.g., four entries), each for storing information regarding a current ATM connection being processed by receive ATM controller 134. Each entry of DMA table 66 includes an identifier of the endpoint 51 in shared memory 44 associated with the connection, identification of the type of cells being processed, and also status flags (error, overflow, etc.) associated with the current packet for that connection. DMA table 66 also includes, for each of its multiple entries, a field for storing partial CRC-32 calculation results for the packet. Reassembly state machine 60 is also in communication with VPI/VCI look-up table 68, which includes an entry that stores, for each valid connection, an index into DMA table 66; as such, reassembly state machine 60 operates in combination with VPI/VCI look-up table 68 to determine whether a received ATM packet is directed to a known connection into host 12. ATM registers 62 are illustrated in FIG. 5 as being in communication with reassembly state machine 60, and include status and control information regarding the reassembly processing.

[0064] VBus state machine 70 is in communication with bus VBus within controller 140 (FIG. 4), and handles both the generation of bus request signals to controller 140, as well as the receipt of ATM data therefrom. The operation of VBus state machine 70 is configured by receive controller VBus registers 72, which are preferably a subset of registers 138 residing on bus VBus as shown in FIG. 4. Word counter 74 and current FIFO address register 76 store the indicated information concerning the source of the received ATM data, which in this example is DSP 32 in modem 14. In particular, current FIFO address register 76 stores a pointer into receive FIFO 33 in the on-chip memory of DSP 32. Word counter register 74 counts the number of data words received in connection with a given packet, so that overflow conditions may be detected and handled by receive ATM controller 134.

[0065] The operation of receive ATM controller 134 according to the preferred embodiment of the invention will now be described in detail relative to FIGS. 6 through 9.

[0066] FIG. 6 illustrates the operation of receive ATM controller 134, and particularly of reassembly state machine 60 therein, in the reassembly of an ATM packet according to the preferred embodiment of the invention. As is well known in the art, ATM transmissions consist of individual cells of fifty-three bytes, with five bytes corresponding to the ATM header of the cell and with the remaining forty-eight bytes available for data (i.e., “payload”); an ATM packet, for example according to AAL5 protocol, in turn is comprised of multiple cells. According to the preferred embodiment of the invention, reassembly state machine 60 operates to effectively remove the ATM header from each received ATM cell, and to direct the payload portion of the received cell to a USB endpoint corresponding to the particular ATM connection indicated in the ATM header; other processing is also performed, as will be described herein.

[0067] The operation of reassembly state machine 60 begins with process 75, in which reassembly state machine 60 causes VBus state machine 70 to poll the status of the receive FIFO buffer 33 resident in DSP 32 of modem 14, and thus to determine whether an ATM cell has been received by DSP 32. Decision 77 is then performed by reassembly state machine 60 to determine whether such a cell is available; if not (decision 77 is NO), poll process 75 is then repeated until such time as a cell becomes available (decision 77 is YES), in response to which control passes to process 78 for the initiation of processing for the ATM cell.

[0068] According to the preferred embodiment of the invention, as evident from FIG. 5, the various state machines 50, 60, 70 of receive ATM controller 134 can operate in parallel with one another to carry out their respective functions. In the case of VBus state machine 70, accesses of DSP 32 via controller 140 and VBUS-to-HPIF bridge 118 (FIG. 4) may be carried out in parallel with the processing of previously received ATM cells by reassembly state machine 60. FIG. 7 illustrates an example of the operation of VBus state machine 70 to access ATM cells from DSP 32 in response to a poll request issued by reassembly state machine 60 in process 75.

[0069] The operation of VBus state machine 70 in response to poll process 75 begins with decision 91 in which VBus state machine 70 polls DSP 32 to determine whether receive FIFO 33 therein is empty. If so (decision 91 is YES), decision 91 repeats. Upon data being stored within receive FIFO 33 of DSP 32 (decision 91 is NO, as receive FIFO 33 is no longer empty), VBus state machine 70 executes process 92 to present the appropriate memory address upon bus VBus, internally within USB interface device 30. As shown in FIG. 7, process 92 builds the Vbus address from the current contents of current FIFO address 76, and includes control values and signals (e.g., length=1, Direction=READ, VBus access request=TRUE, etc.,); upon grant of VBus access by controller 140 to VBus state machine 70, VBus state machine 70 applies the VBus address and control signals to bus VBus, following which VBUS-to-HPIF bridge 118 converts the VBus address and signals into the appropriate address for accessing DSP 32. VBus state machine 70 then executes decision 93 to determine whether bus VBus indicates that the read to DSP 32 is ready; if the read has not yet been accomplished (decision 93 is NO), VBus state machine 70 continues to wait by reexecuting decision 93. Upon the completion of the read by DSP 32 applying the contents of receive FIFO 33, as translated by VBUS-to-HPIF bridge 118 onto bus VBus (decision 93 is YES), VBus state machine 70 receives the data word then present on bus VBus and stores it into a buffer location for forwarding to reassembly state machine 60 as requested, in process 94. The contents of current FIFO address register 76 is then incremented, in process 96, and control passes back to decision 91 for polling of the state of receive FIFO 33 of DSP 32 as noted above.

[0070] As noted above, the data retrieval operation performed by VBus state machine 70 is performed in parallel with the continued operation of reassembly state machine 60 such as illustrated in FIG. 6. This parallel operation permits the forwarding and processing of data in the manner described herein, while reducing the extent to which the operation stalls to await data. Additionally, as will be noted below, this operation permits the forwarding of a leading portion of an ATM data packet to host 12 even before the entire packet has been retrieved from DSP 32.

[0071] Referring back to FIG. 6, upon reassembly state machine 60 determining that an ATM cell is available (decision 77 is YES), reassembly state machine 60 then performs process 78 to retrieve the first four bytes of the new ATM cell. As is known in the art, these first four bytes of the five byte ATM header include connection information presented in the VPI/VCI field, as well as an indication of the type of ATM cell that is received. In process 78, reassembly state machine 60 interrogates the information in these first four bytes in order to carry out a series of decisions and operations for processing the received cell, as will now be described.

[0072] In decision 79, reassembly state machine 60 determines whether the received cell is an OAM cell. As is known in the art, an OAM cell is an “Operation and Maintenance” cell, and constitutes a single-cell message packet for communication of maintenance and configuration information regarding the ATM connection. If the received cell is an OAM cell (decision 79 is YES), control passes to process 80 in which reassembly state machine 60 processes the OAM cell, which will now be described with reference to FIG. 8. In this regard, it should be noted that reassembly state machine 60 may be configured to respond to decision 79 only in connection with specific classes of DAM cells. For example, only OAM cells of only one of the F4 or F5 flow types may be processed by process 80; alternatively, OAM cells of either class may be processed and forwarded in the manner described below.

[0073] According to the preferred embodiment of the invention, the execution of process 80 begins with process 200, in which the remainder of the OAM ATM cell (specifically, the remaining forty-nine bytes) is fetched. As noted above relative to FIG. 7 and the description of the operation of VBus state machine 70, the fetching of cell contents is preferably performed in parallel with the operation of reassembly state machine 60, and as such is not a separate sequentially-performed process in the strict sense, as may be suggested by FIG. 8. In any event, once the current ATM cell is received through process 200, reassembly state machine 60 and receive state machine 50 forward the received cell payload to filter endpoint 51₁of shared memory 44 in process 202; the retrieved ATM header need not be forwarded, as its information has been utilized in the determination of the cell type (OAM) and also the destination (filter endpoint 51₁), and is no longer necessary. In addition to forwarding the received cell payload, CRC logic 64 performs a ten-bit cyclic redundancy check (CRC) on the received cell payload and, in decision 203, reassembly state machine 60 evaluates the result of the CRC-10-operation. If an error is detected (decision 203 is NO), reassembly state machine 60 sets a CRC error bit in process 204. In either case (i.e., decision 203 is YES, or after process 204 if decision 203 is NO), reassembly state machine 60 then performs process 206 to generate an end-of-packet (EOP) notification to host 12. As discussed above, such notifications are generally performed by way of a write to completion endpoint 51₀in shared memory 44, In this case, the state of the CRC error bit is forwarded to host 12 via completion endpoint 51₀, indicating the validity of the OAM cell previously forwarded. Processing of the OAM cell is then completed, and control passes back to polling process 75 (FIG. 6) to await the next received ATM cell. Host 12 will decode and respond to the OAM cell as appropriate; according to this preferred embodiment of the invention, no ATM maintenance functions are carried out by USB interface device 30.

[0074] As noted above, receive state machine 50 communicates with the appropriate endpoints 51 in shared memory 44 over bus B, independently from and simultaneously with the processing being carried out by reassembly state machine 60. In processes 202, 206 noted above, as well as in the other forwarding processes noted hereinbelow, receive state machine 50 operates in conjunction with multiplexer 106 and buffer byte counter 55 to receive a complete data word (i.e., the four bytes of byte buffers 58), and to forward the same to the appropriate buffer location indicated by the contents of current buffer pointer 54, at the memory address for the corresponding pointer location as provided in configuration registers 52. It is contemplated that those skilled in the art having reference to this specification will be readily able to implement this function by way of sequential logic, as noted above.

[0075] Referring back to FIG. 6, if the received cell is not an OAM cell (decision 79 is NO), the received cell then likely corresponds to an ATM cell that is part of a communication packet. Reassembly state machine 60 then performs decision 81 to evaluate whether the retrieved ATM header information indicates that the cell belongs to an ATM connection (virtual connection) that is known by USB interface device 30 and host 12. The evaluation of decision 81 is performed by reassembly state machine 60 applying the corresponding VPI/VCI fields from the ATM cell header to VPI/VCI look-up table 68; if the connection is unknown, no valid index will be returned from look-up table 68 and the result of decision 81 will be NO. In this event, reassembly state machine 60 will write the value of the received VPI/VCI fields from the ATM cell header to one of ATM registers 62, specifically a receive-unknown register therein in process 82, will increment a counter corresponding to unknown protocol (such counter also preferably resident in ATM registers 62), and will initiate a flush of the remainder of this unknown connection cell from receive FIFO 33 in DSP 32. Control then passes back to polling process 75 to await the next ATM cell.

[0076] If look-up table 68 returns a valid index, the connection is known and decision 81 is YES. Reassembly state machine 68 then performs process 86 to retrieve the connection information for the VPI/VCI value in the received ATM header. As noted above, ATM receive controller 134 is capable of simultaneously supporting three ATM virtual connections, via three endpoints 51₂through 51₄As such, successive cells processed by reassembly state machine 60 will generally not be associated with the same packet. Accordingly, each instance of polling process 75 will simply be interrogating receive FIFO 33 in DSP 32 for any cell, rather than a received cell for a particular packet, with decision 81 and process 86 serving to associated the current cell with the correct packet and virtual connection. According to the preferred embodiment of the invention, process 86 is performed by reassembly state machine 60 retrieving endpoint information from the location of DMA table 66 to which the returned index from VPI/VCI look-up table 68 points. The retrieved connection information includes information regarding the type of packet for the connection to which the received cell belongs, the one of endpoints 51 in shared memory 44 to which the received cell is to be forwarded, as well as status information regarding the cells of the corresponding ATM packet that have been already received and processed (such status information including the length of the ATM packet so far, error and status bits, and also the partial CRC results so far). Following the retrieval of this information, reassembly state machine 60 executes process 88 to process the payload portion of the current cell, as will now be described relative to FIG. 9.

[0077] Process 88 begins with the fetching of the remainder of the received ATM cell from receive FIFO 33 of DSP 32, as performed by VBus state machine 70 in conjunction with VBUS-to-HPIF bridge 118 in process 208; as noted above, process 208 may be performed in parallel with the operations of reassembly state machine 60, and as such need not necessarily be performed as a sequential process. Reassembly state machine 60 then determines, in decision 209, whether the current received cell is the first cell of an ATM packet. As is known in the art, the first cell of an ATM packet is not expressly, indicated in the ATM cell header. Rather, it is contemplated that decision 209 is performed by testing a flag or other internal status indicator that is set in connection with the completion of the only or last cell of a previous ATM packet. For example, if the previously processed cell for the current connection was the cell at the end of a packet, this flag would be set to indicate that the next cell to be processed is necessarily the first cell of the next packet for this connection, permitting reassembly state machine 60 to properly determine whether the current cell is the first one. Alternatively, various packet data as stored in DMA table 66 for the connection may be cleared upon completion of a packet, in which case reassembly state machine 60 can determine that the current received cell is the first cell of a packet, by virtue of the invalid status information for that connection.

[0078] In any event, for the first cell of a packet (decision 209 is YES), reassembly state machine 60 controls CRC logic 64 to begin the CRC-32 calculation for the packet with which the current cell is associated. As known in the art, multiple-cell ATM packets, such as those corresponding to the AAL5 protocol, include a thirty-two bit cyclic redundancy check sum that is calculated over the payload of the entire packet. According to the preferred embodiment of the invention, CRC logic 64 in ATM receive controller 134 is dedicated hardware for calculating this CRC checksum (as well as the ten-bit CRC value used for single-cell OAM packets, discussed above). The partial results of CRC calculation process 210 are stored in the entry of DMA table 66 associated with the current connection. This hardware calculation is performed in parallel with the other processing of reassembly state machine 60 in processing the cells of a packet, rather than in host software as conventionally performed; as is known in the art, CRC calculations performed by general purpose programmable devices is a relatively complex process, particularly over long data streams such as multiple-cell ATM payloads. This CRC check performed in hardware according to the preferred embodiment of the invention is thus significantly more efficient than these conventional software-based approaches.

[0079] The processing of the first cell of the packet continues with process 212, in which the entry of DMA table 66 corresponding to the cell count of the associated packet is set to “1”, indicating that one cell has been received so far for this packet. In process 214, reassembly state machine 60 and receive state machine 50 forward the payload of the current received cell to the bulk endpoint 51₂through 51₄, indicated by the connection information stated in DMA table 66, as indexed by VPI/VCI look-up table 68. As noted above, the forwarding of the payload to this endpoint 51 is performed by receive state machine 50 being signaled that a full data word is ready in byte buffers 58, following which receive state machine 50 obtains access to bus B and writes the data to the corresponding endpoint in shared memory 44.

[0080] As is evident from the foregoing description, these bytes correspond to the payload portion of the ATM cells, and do not include the header portion of the ATM cells, which is effectively stripped by reassembly state machine 60 (i.e., the ATM cell header is simply not forwarded by reassembly state machine 60 to byte buffers 58). Additionally, boundaries between ATM cells are effectively ignored in the writing of the payload to shared memory 44, such that the payload portion of successive ATM cells in the same ATM packet are contiguously stored in shared memory 44 at the endpoint 51 corresponding to the virtual connection. These contiguous cell payloads are thus contiguously retrieved from shared memory 44 by host 12 over the USB bus.

[0081] As noted above, this forwarding to shared memory 44 can also be performed in parallel with the other operations of ATM receive controller 134, including the fetching of additional cell data from DSP 32 and the processing of cells by reassembly state machine 60, including the CRC calculation. Control then passes back to polling process 75 to await the receipt of the next ATM cell.

[0082] If the current received cell is not the first cell of a packet (decision 209 is NO), reassembly state machine 60 then executes decision 211 to determine whether the received cell is the cell at the end of a packet. This determination is made by examination of the ATM header for the current received cell, as is known in the art. If the received packet is not an end-of-packet cell (decision 211 is NO), control then passes to process 216, in which CRC logic 64 is directed, by reassembly state machine 60, to continue the CRC-32 calculation with the payload data fetched in connection with the current cell. This calculation begins with the partial CRC results for the associated connection, retrieved from DMA table 66, and upon the completion of the partial CRC calculation for the current cell payload, this entry of DMA table 66 is then rewritten with the updated value of these intermediate results. The cell count in the corresponding entry of DMA table 66 with the current connection is incremented to reflect the processing of another cell, in process 218. The cell payload (i.e., without the ATM header) is then forwarded to the appropriate endpoint 51, in process 214 as described above.

[0083] Upon reaching the last cell in the packet (decision 211 is YES), reassembly state machine 60 then executes various steps to complete the forwarding of the packet to host 12. According to this embodiment of the invention, a YES result returned by decision 211 passes control to process 220, in which the cell count in the corresponding entry of DMA table 66 is incremented to reflect the receipt of another (and final) cell associated with the current packet for the corresponding connection. This final cell count value will be used in generating the end-of-packet notification, described hereinbelow. In addition, this final cell count value is retrieved by reassembly state machine 60 and compared, in decision 221, against a maximum cell count. In this exemplary embodiment of the invention, the maximum cell count is 1366₁₀, which corresponds to the 64 k byte maximum length of an ATM packet according to current standards. If the number of received cells exceeds this maximum length, one may conclude that a previous end-of-packet indicator was not detected, because of a transmission error or some other reason. In this event, some of the previously received cells that have been associated with the current packet may not properly be part of the current packet. According to the preferred embodiment of the invention, reassembly state machine 60 detects such an event by decision 221 detecting that the cell count exceeds the maximum length limit (a YES result), and by executing process 222 to set a packet overflow bit in the corresponding entry of DMA table 66 for this packet.

[0084] In either case (decision 221 YES or NO), reassembly state machine 60 next executes decision 225 to determine whether the current packet is an AAL5 protocol packet; this determination is made by interrogating the entry of DMA table 66 corresponding to the current packet. If the current packet is not an AAL5 packet (decision 225 is NO), control passes to process 226 for completion of the processing of the current packet by forwarding of the payload of the last cell to the appropriate endpoint 51 in process 226 (performed in cooperation with receive state machine 50 as described above). In this case, as in the case of previously-received cells, the ATM header information for this cell is not forwarded to the endpoint. Additionally, reassembly state machine 60 performs process 228 to generate an end-of-packet notification to host 12. This notification, similarly as described above relative to FIG. 7 for the case of OAM cell processing, involves the writing of data to endpoint 51₀in shared memory 44, including information regarding the packet (including at least the state of the packet overflow bit, which may have been set in process 22). The processing of this non-AAL5 packet, in this example, is then complete.

[0085] In the case of an AAL5 packet (decision 225 is YES), as typically utilized in modern ATM communications, certain trailer information is received, for analysis and forwarding following the final cell payload. Decision 227 is next executed by reassembly state machine 60 to determine if the packet length value in the AAL5 trailer has the value “0”. If so (decision 227 is YES), an error in the packet was encountered somewhere along its transmission; reassembly state machine 60 then sets an AAL5 abort bit in the status information of the entry of DMA table 66 corresponding to the current packet. Control then passes to process 226 for forwarding of the cell payload to the appropriate endpoint 51, and to process 228 for generation of the EOP notification to host 12 (including the state of the AAL5 abort bit, as set in process 230).

[0086] If the length value of the AAL5 trailer does not indicate an abort code (decision 227 is NO), reassembly state machine 60 will then execute decision 231 to evaluate the result of the CRC-32 calculation carried out by CRC logic 64. Upon receipt and fetching of the last cell in the packet, CRC logic 64 will have calculated the entire CRC-32 value over the payload of the multiple cells in the packet, In decision 231, reassembly state machine 60 compares the result of the CRC calculation by CRC logic 64 with the expected (i.e., valid) value communicated as part of the last four bytes of the received cell. If the two values do not match (decision 231 is NO), an error is present in the received packet, and reassembly state machine 60 sets the CRC error bit in the entry of DMA table 66 corresponding to the current packet, in process 232.

[0087] In either case (decision 231 YES or NO), control passes to process 226 for the forwarding of the cell payload to the appropriate endpoint 51 in shared memory 44, as noted above. Reassembly state machine 60 then executes process 228 to generate the end-of-packet notification, for forwarding to completion endpoint 51₀. This EOP notification includes various status information contained within the corresponding entry of DMA table 66 for this packet, including the clear state of the AAL5 abort bit, the state of the CRC error bit (as may have been set in process 232), and of course an indication that a complete packet has now been forwarded to the endpoint 51 associated with the current virtual connection.

[0088] Alternatively, an end-of-packet completion notification may be appended to the payload forwarded to the appropriate endpoint 51₁through 51₄, in which case a separate completion endpoint 51₀need not be maintained within shared memory 44.

[0089] Once the EOP notification has been written to completion endpoint 51₀, host 12 will be notified, during its next poll of this bulk endpoint, that cells for an ATM packet are now stored in shared memory 44, at the appropriate IN endpoint 51. Host 12 may then begin reading the payload data stored at endpoint 51, over the USB bus, in the form of USB packets. According to current standards, USB packets for communication with bulk endpoints are sixty-four byte packets. Referring to FIG. 3, serial interface engine 114 then effects the communication of the contents of RAM 106 (i.e., shared memory 44 of FIG. 2) to host 12 over bus USB. Because of the arrangement of the cell payloads in shared memory 44 (or RAM 106) as a contiguous data block, without ATM headers, the fifty-three byte format of ATM cells need not be followed in the communication of the payload data to host 12 over bus USB; instead, the data may be forwarded in full sixty-four byte USB packets.

[0090] As is evident from the foregoing description, the construction and operation of ATM receive controller 134 within USB interface device 30 provides important advantages in the efficiency of ATM communications. One such advantage is a significant improvement in the efficiency and utilization of the USB bus in the communication of the ATM data. The processing performed by ATM receive controller 134, particularly in associating the ATM packet with a specific endpoint, permits the five-byte ATM header to be effectively stripped from the ATM cells prior to forwarding over bus USB, thus reducing the amount of overhead transmitted over bus USB and improving the bus efficiency by on the order of ten percent. Improved bus utilization is also provided by the ability to use the entire maximum sixty-four byte USB packet for payload communication, as the ATM cell boundaries need not be maintained over the USB bus because of the processing provided by ATM receive controller 134; in contrast, conventional systems in which the fifty-three byte ATM cells are communicated over the USB bus necessarily include eleven vacant bytes within each USB packet (indeed, sixteen bytes having no payload), in order to preserve the USB cell boundaries. Furthermore, the amount of memory required in the USB interface device is greatly reduced from that in conventional interface devices, as the ATM packets are reassembled directly into the memory of the host via the USB endpoints. As a result, the shared memory in the USB interface device need not be sufficiently large to store an entire maximum-size ATM packet; rather, the packet reassembly may be fully handled through the endpoint FIFO buffers. Processing efficiency is also provided by way of the present invention, since computationally intensive operations such as CRC calculation may be performed in hardware in the USB interface device, rather than in software at the host, which is necessarily less efficient and cumbersome; the offloading of the reassembly function to the USB interface device also dramatically simplifies the development and installation of the ATM communications driver at the host.

[0091] ATM Cell Segmentation

[0092] USB-based DSL modem 14 in system 10 according to the preferred embodiment of the invention not only handles received ATM packets, but of course also provides the functionality of transmission, in ATM packets, of data generated by host 12. According to the preferred embodiment of the invention, much of the processing required for the segmentation of ATM packets is carried out in USB interface device 30, thus relieving host 12 from performing a significant portion of the computationally intensive segmentation operations, as now executed by host software in conventional USB-based DSL modems. Specifically, it is contemplated that USB interface device 30 efficiently performs the ATM segmentation of sixty-four byte USB packets into the fifty-three byte standard ATM cells, including generation of the ATM header for each cell; USB interface device 30 also receives and forwards the necessary “padding” of the ATM cells to fill the fifty-three byte cells, and also calculates and appends of the appropriate cyclic redundancy check (CRC) value to the packet. USB communication is thus made more efficient by not requiring the host to forward each copy of the ATM header over the USB bus, and by concatenating payload data into the maximum size sixty-four byte USB packets, thus greatly reducing the amount of null bytes over the USB bus.

[0093] Referring now to FIG. 10, the construction of ATM transmit controller 132 according to the preferred embodiment of the invention will now be described. As shown in FIG. 10, ATM transmit controller 132 is primarily controlled by state machines, namely transmit state machine 250, segmentation state machine 260, and VBus state machine 270, each of which is preferably implemented by sequential logic. It is contemplated that those skilled in the art, having reference to this specification, will be readily able to implement state machines 250, 260, 270 to perform the functions described herein, and in a manner suitable for particular realizations.

[0094] Transmit state machine 250 is coupled to bus B in USB interface device 30, for controlling communication with transmit endpoint 240 in shared memory 44 (which, in the exemplary realization of FIG. 3, resides in RAM 106). According to the preferred embodiment of the present invention, both ATM header information and also the ATM packet payload are written into transmit endpoint 240 by host 12; alternatively, separate header and payload endpoints may be established in shared memory 44, if desired. According to this embodiment of the invention, transmit endpoint 240 is a bulk USB endpoint, implemented by one or more sixty-four byte buffers in a conventional FIFO manner (e.g., a pair of “Ping-Pong” buffers or an elastic FIFO arrangement). Transmit state machine 250 is in communication with endpoint configuration register 242, which is written by MCU 100 with the memory address of endpoint 240 in shared memory 44. Additionally, transmit state machine 250 is controlled by the contents of current buffer pointer 244, which indicates the FIFO buffer at endpoint 240 from which data is being retrieved, and by buffer byte counter 246 which indicates the byte position within the accessed FIFO buffer, as used in handshaking operations with USB buffer management circuitry, as noted above.

[0095] Segmentation state machine 260 receives data bytes from transmit state machine 250, and performs segmentation operations as appropriate for the particular packet upon such bytes. The bytes processed by segmentation state machine 260 are forwarded, by way of demultiplexer 266, to the appropriate one of byte buffers 268. The operation of segmentation state machine 260, which will be described in further detail hereinbelow, is carried out in combination with header register 252, packet length register 254, and packet type register 256, each of which stores results useful in the generation of ATM headers. Segmentation state machine 250 also operates in cooperation with CRC logic 258, which performs CRC calculations to generate the CRC filler of the transmitted ATM packets. Additionally, cell counter 262 counts the number of ATM cells being generated by segmentation state machine 260 for the current packet, and byte counter 264 counts the number of bytes within the current ATM cell that have been processed by segmentation state machine 250.

[0096] As noted above, demultiplexer 266 receives each byte processed by segmentation state machine 260, and forwards it to the appropriate one of byte buffers 268. The contents of byte registers 268 are simultaneously received, as a thirty-two bit data word, by VBus state machine 270. In combination with word counter 274 and transmit configuration registers 272, VBus state machine 270 operates to apply the data word to bus VBus, and thus to transmit FIFO 35 in DSP 32 (FIG. 2) over bus VBus and via VBUS-to-HPIF bridge 118 (if present, as in FIG. 3).

[0097] The operation of transmitting ATM communications begins, of course, with host 12 generating the message to be transmitted. According to the preferred embodiment of the invention, host 12 generates the data packet from its own data processing operations, and also generates an ATM header and appropriate control information for the overall packet; as segmentation is not performed by host 12 according to this embodiment of the invention, the message data is not segmented into ATM cells by host 12, nor is an ATM header generated for each such ATM cell. Host 12 formats this packet into sixty-four byte USB packets, and transmits the USB packet containing the ATM header and control information over the USB bus to transmit endpoint 240 in shared memory 44, followed in the same packet by the payload, or message data, which are also written to transmit endpoint 240. Host 12 will, of course, control the scheduling of these USB packets over the USB. Furthermore, to avoid error according to the preferred embodiment of the invention, host 12 is to communicate only one ATM packet over the USB bus at a time, and ATM transmit controller 132 is to process that one ATM packet, (i.e., only the one virtual connection) at a time; it is of course contemplated that multiple instances of ATM transmit controller 132, and associated endpoints 240, may be implemented in order to handle multiple connections, if desired.

[0098] FIG. 11 illustrates the arrangement of the packet formats (i.e., packet definition units, or PDUs) that may be transmitted from host 12 to USB interface device 30, and that may be segmented by ATM transmit controller 132 according to the preferred embodiment of the present invention. AAL5 PDU 230, as shown in FIG. 11, includes a seven-byte header, followed by up to 64 k bytes of payload data. A pad portion of up to 47 bytes follows the payload, and simply contains a number of null bytes sufficient to fill out a forty-eight byte boundary, including a two-byte control field, a two-byte length field, and a four byte CRC-32 filler that all follow the payload, and not including the seven-byte header. OAM PDU 232 includes a seven-byte header, followed by payload data of forty-six bytes and six bits, which in turn is followed by the ten-bit CRC-10 checksum value, These AAL5 and OAM cells are well-known in the art.

[0099] Two other PDUs that may be handled by ATM transmit controller 132 according to the preferred embodiment of the invention include a PTI-based Null-AAL shown as PDU 234 of FIG. 11, and transparent packet PDU 236 which is useful in generic streaming applications. According to this embodiment of the invention, PTI Null PDU 234 consists simply of a seven-byte header, and up to 64 k bytes of payload data, with up to forty-seven bytes of pad to fill out the forty-eight byte boundaries; as such, PTI Null PDU 234 is sufficiently generic to be used, in connection with downstream software, to support AAL1, AAL2, and AAL3/4 protocols. Transparent PDU 236 is simply a fifty-three byte packet with seven bytes of header and forty-eight bytes of data, similar to a conventional ATM cell.

[0100] Referring now to FIG. 12, the operation of ATM transmit controller 132, and particularly segmentation state machine 260 therein, in performing the segmentation of the ATM packet generated by host 12 into ATM cells for transmission via modem 14. The operation of ATM transmit controller 132 begins with decision 281, in which segmentation state machine 260 interrogates the state of byte buffers 268 to determine whether a location is available to which to forward another byte of packet data. If not (decision 281 is YES because byte buffers 268 are full), segmentation state machine 260 must wait until VBus state machine 270 reads the data word from the contents of byte buffers 268, following which decision 281 will return a NO result, permitting control to pass to decision 283.

[0101] As noted above, VBus state machine 270 writes the transmit data words to transmit FIFO 35 of DSP 32 independently from, and thus simultaneously with, the operation of segmentation state machine 260 described herein. This operation by VBus state machine 270 consists of requesting access to bus VBus and, upon receiving a grant of such access, reading the contents of byte buffers 268 as a transmit data word, and then presenting the transmit data word to bus VBus; as such, VBus state machine 270 operates substantially similarly with VBus state machine 70 of ATM receive controller 134 described hereinabove, except writing data to bus VBus rather than reading data therefrom. Similarly as described above, VBus state machine 70 maintains the current transmit FIFO address in its current FIFO address 276, so that the proper memory address may be presented in combination with the data word to be transmitted. VBus-to-HPIF bridge 118 then translates, or “bridges”, this address and data information into the HPIF format comprehendible by DSP 32. Once VBus state machine 270 clears byte buffers 268, as noted above, decision 281 will return a NO result to indicate that space is available in byte buffers 268 for the next byte to be processed.

[0102] In decision 283, segmentation state machine 260 in combination with transmit state machine 250 determine whether new data is present at endpoints 240, either at header endpoint 240₀or data endpoint 240₁. If not (decision 283 is NO), segmentation state machine 260 must wait for such data to appear. Upon data being written to endpoints 240 from host 12 for an ATM packet, decision 283 will return a YES result, and segmentation state machine 260 proceeds to execution of decision 285.

[0103] In decision 285, segmentation state machine 260 determines whether the ATM header for the current packet has yet been configured. Upon receiving the initial packet information, typically the ATM header and control information at endpoint 240₀, the ATM header will not have yet been configured. In this case (decision 285 is NO), segmentation state machine 260 copies the received ATM header and control information into its registers 252, 254, 256, shown in FIG. 10. Specifically, the first four bytes of the seven-byte ATM header information read from endpoint 240₀, corresponding to the GFC, VPI, and VCI connection information, are stored in header register 252, the fifth and sixth bytes corresponding to the length of the packet are written into length register 254, and the seventh byte corresponding to the packet type is written into packet type register 256. Following the storing of this information in registers 252, 254, 256, segmentation state machine 260 retains the information necessary for generation of the five-byte ATM cell headers for each ATM cell segmented and forwarded by ATM transmit controller 132. Control then passes to process 288, for transmission of header bytes to DSP 32, as will now be described relative to FIG. 13.

[0104] As noted above and as well known in the art, conventional ATM cell headers are five bytes in length; however, only a thirty-two bit data word is communicated over bus VBus to bus HPIF, and thus to DSP 32. As such, the transmission of an ATM cell header requires two data words for complete transmission. However, as is also well known in the art, the fifth byte of an ATM cell header is dedicated for the HEC field, which will be generated by the transmission convergence layer at DSP 32. As such, while segmentation state machine 260 must generate a five byte ATM cell header, the fifth byte of this header is a null, or dont care, byte, prior to processing by DSP 32. Process 288 is thus a two-pass process, and begins with decision 311 in which segmentation state machine 260 determines whether the four-byte partial header has previously been transmitted to VBus state machine 270. If not (decision 311 is NO), control passes to decision 313.

[0105] In decision 313, segmentation state machine 260 determines, from the packet type ATM header information stored in packet type register 256, whether the current packet is an OAM packet. If so (decision 313 is YES), the transmitted packet will be a single cell packet, and as such control passes directly to process 314 by way of which segmentation state machine 260 transfers the first four bytes of the ATM cell header to byte buffers 268 via demultiplexer 266. As is known in the ATM art, these four bytes include the GFC, VPI, VCI, PTI, and CLP fields. Control then passes back to decisions 281 and 283 to wait for byte buffers 268 to clear and for additional data to be written to endpoint 240_1.

[0106] If decision 313 is not an OAM or a transparent packet (decision 313 is NO), segmentation state machine 260 performs decision 315 to determine whether the last ATM cell is about to be transmitted for the current ATM packet. If so (decision 315 is YES), process 316 is performed by segmentation state machine 260 to rewrite the PTI field in the four-byte ATM header to include the end-of-packet indicator for that cell. In either case (decision 315 is NO, or following process 316), segmentation state machine 260 transfers the first four bytes of the ATM cell header to byte buffers 268, in process 314.

[0107] Following process 314, control passes back to decision 281 to await the reading of the previous header information from byte buffers 268. Once byte buffers 268 are read, and upon decision 283 returning a YES, indicating the presence of new data at endpoint 240₁decision 285 again returns a NO result since the header has not been completely configured and transmitted. Process 286 may be skipped for this second pass (considering that the ATM header information is already stored in registers 252, 254, 256) and control passes to decision 311 of process 288 which returns a YES result indicating that the four-byte partial header has already been set. Segmentation state machine 260 then transmits a null byte to one of byte buffers 268 (i.e., the zeroth byte) for eventual transmission to DSP 32. Control then returns back to decision 281 (which necessarily returns a NO result at this point), and decisions 283, 285 (each of which returns a YES result), following which segmentation state machine 260 next performs decision 287.

[0108] Decision 287 determines whether the current byte count value, which is stored in byte counter 264, has the value forty-eight, which is the maximum number of bytes of payload in an ATM cell, as is known in the art. If not (decision 287 is NO), the next byte of data to be processed will not be the final byte in the current ATM cell. Decision 293 is then performed by segmentation state machine 260 to determine whether the current ATM cell is the last cell in the ATM packet, by comparing the current value of cell counter 262 with a terminal cell count that may be derived from the current contents of length register 254; in this regard, since all OAM packets consist of only a single cell, a packet type identifier (register 256) indicating an OAM packet will also cause decision 293 to return a YES result.

[0109] If the current ATM cell being produced is not the last cell in the packet, process 294 is performed to cause transmit state machine 250 to fetch the next data byte from endpoint 240₁of shared memory 44; this next data byte is received by segmentation state machine 260, and CRC logic 258 initiates or continues the calculation of the CRC value for the current ATM packet. As discussed above relative to the receive and reassembly of ATM packets, AAL5 ATM packets include a trailer containing a thirty-two bit CRC checksum for the payload of all cells within the ATM packet. CRC logic 258 provides dedicated hardware for the calculation of this CRC checksum in an ongoing manner, effectively in parallel with the processing carried out by segmentation state machine 260. Following the fetch of the next data byte (but not necessarily the completion of the CRC calculation, as noted above), segmentation state machine 260 next increments the value of byte counter 264, in process 296, and then transfers the fetched data byte to the next open location in byte buffers 268. As noted above, VBus state machine 270 retrieves the contents of all four byte buffers 268, in the form of a thirty-two bit data word, once these buffers are filled. Whether read by VBus state machine 270 or not, after the transfer of the data byte to byte buffers 268, control passes back to decision 281, where segmentation state machine 260 determines whether an available byte buffer remains, and then whether data is present at endpoint 240₁, as noted above.

[0110] If the current cell is the last cell in the ATM packet (i.e., decision 293 is YES), additional processing is required to generate trailing information, or at least to prepare ATM transmit controller 132 for the next packet. Decision 299 is first performed to determine whether the current cell corresponds to an AAL5 or OAM packet; if it is of neither type (decision 299 is NO), process 302 is then performed to simply fetch the next data byte from endpoint 2401. If, on the other hand, the current packet is an AAL5 or OAM packet, process 300 is performed by segmentation state machine 260 to substitute the final CRC value (either CRC-32 or CRC-10, as the case may be) for the appropriate data portion of the cell. In either case (after process 302 or process 304, as the case may be), decision 303 is next performed to determine whether the current byte is the last byte in this the last cell of the packet, If not (decision 303 is NO), process 296 is performed to increment the byte count and the current byte is transferred to the appropriate byte buffer 268 in process 298, with control returning to decision 281 for receipt of the next byte.

[0111] Upon reaching the last byte of the cell (decision 303 is YES), the ATM packet may be completed. In process 304, the header and control configuration information stored in registers 252, 254, 256 is cleared by segmentation state machine 260, so that the next packet will have its own ATM cell headers configured and transmitted via decision 285 and processes 286, 288. The contents of byte counter 264 are cleared in process 306, and the contents of cell counter 262 are cleared in process 308. This final byte is then transferred to byte buffers 268, in process 298, and the transmission of the ATM packet is then complete. Control then returns to decisions 281, 283 to await the emptying of byte buffers 268 and the receipt of new data at endpoints 240, with the process then repeating for the next ATM packet.

[0112] According to the present invention, therefore, significant advantages are obtained by the segmentation of ATM packets into ATM cells at the USB interface device, as described above. The USB bus is more efficiently utilized according to the present invention, as compared with conventional USB-based devices, because the ATM header information need only be communicated over the USB bus once according to the present invention; the segmentation logic in the USB interface device itself then generates the ATM cell headers. Secondly, the host is able to communicate the ATM packet payload by way of full USB packets (e.g., sixty-four byte bulk packets), without regard to ATM cell boundaries, as opposed to conventional devices in which the fifty-three byte ATM packets are transmitted within individual sixty-four byte USB packets with eleven null bytes. The computationally intensive operations of segmentation, CRC calculation, and the like are also performed in the USB peripheral according to the present invention, preferably in dedicated hardware, thus relieving the host from performing these functions in software. Accordingly, the development and implementation of host drivers for the ATM transmission is significantly facilitated by the present invention.

[0113] While the present invention has been described according to its preferred embodiments, it is of course contemplated that modifications of, and alternatives to, these embodiments, such modifications and alternatives obtaining the advantages and benefits of this invention, will be apparent to those of ordinary skill in the art having reference to this specification and its drawings. It is contemplated that such modifications and alternatives are within the scope of this invention as subsequently claimed herein.