Real-Time Disk System

1. Field of the Invention

The present invention relates generally to a system that can record or play short segments of digital component video on specially-modified computer disk storage media. More particularly, it relates to such a system in which the video segments are stored and retrieved directly in parallel from the disk storage media without serial-to-parallel or parallel-to-serial conversion of a video signal stream. It further relates to such a system that can be expanded two dimensionally for multiuser and larger capacity requirements. It further relates to such a system incorporating smoothed motion.

2. Description of the Prior Art

It is known to record video on magnetic disks in order to be able to retrieve and display stored video images in real time. Commercially available real-time disk systems are available from Abekas and Quantel. The Quantel product is described in U.S. Pat. No. 4,668,106, issued Aug. 18, 1987 to Keller et al. The system disclosed by Keller et al. uses parallel-transfer disks to record 4:2:2 D1 digital video images. However, the number of parallel data channels on the disk does not match the number of bits in a pixel. A complicated parallel to serial converter is therefore required to record on disk. U.S. Pat. Nos. 4,638,381; 4,647,986 and 4,674,064, issued Jan. 20, 1987, Mar. 3, 1987 and Aug. 18, 1987 to Vaughn, Vaughn et al. and Vaughn disclose a parallel-transfer disk system for real-time recording of digitized X-rays, but this system also does not have the same number of parallel data channels on the disk as the number of bits in a pixel. It therefore also requires a very complicated serial-to-parallel and parallel-to-serial converter.

A system for generating interlaced slow motion video by spatial and temporal interpolation is described in U.S. Pat. No. 4,987,489, issued Jan. 22, 1991 to Hurley et al. In this system, successive fields of an input video signal are stored in field stores and are spatially interpolated as well as temporally filtered to produce new fields depending on the amount of motion detected in a scene.

Accordingly, it is an object of this invention to provide a real-time disk system in which video images are stored on a disk and retrieved from the disk in parallel signal streams without requiring any parallel to serial or serial to parallel conversion.

It is another object of the invention to provide such a real-time disk system that can be expanded two dimensionally for multiuser and larger capacity requirements.

It is a further object of the invention to provide such a system incorporating smoothed slow motion utilizing motion-adaptive temporal-linear interpolation and frame repetition to produce a smooth fade over between two frames.

It is still another object of the invention to provide a video processing system with smoothed slow motion which is able to perform film-to-video transfers.

It is a still further object of the invention to provide such a video processing system with smoothed slow motion which provides film-to-video transfers with reduced jitter and judder artifacts.

The attainment of these and related objects may be achieved through use of the novel real-time disk system herein disclosed. A real-time disk system in accordance with this invention has a video processor connected by a plurality of parallel data channels to a disk storage means having a like plurality of storage surfaces and a like plurality of interface circuits. One of the like plurality of interface circuits is connected between one of the plurality of parallel data channels and one of the like plurality of storage surfaces.

An improved video processing system in accordance with the invention has a means for smooth motion processing a group of video fields by creating a plurality of additional fields between two original fields in the group of video fields by a combination of motion adaptive interpolation and frame repetition.

The attainment of the foregoing and related objects, advantages and features of the invention should be more readily apparent to those skilled in the art, after review of the following more detailed description of the invention, taken together with the drawings, in which:

FIG. 1 is a block diagram of a real-time disk system in accordance with the invention.

FIG. 2 is a more detailed block diagram of a first portion of the system shown in FIG. 1.

FIG. 3 is a more detailed block diagram of a second portion of the system shown in FIG. 1.

FIG. 4 is a more detailed block diagram of a third portion of the system shown in FIG. 1.

FIGS. 5 and 6 are flow charts showing operation of the system of Figures 1-4 for recording and playing back video images.

FIG. 7 is a schematic representation useful for understanding operation of a portion of the system shown in FIGS. 1-5.

FIG. 8 is a block diagram of a fourth portion of the system shown in FIG. 1.

FIGS. 9A and 9B are block diagrams of another embodiment of the system portion shown in FIG. 8.

FIG. 10A is a block and flow diagram representation of operation of the system in a Smooth Motion Mode.

FIG. 11 is a conceptual block and schematic diagram integrating the various modes of the system shown in FIGS. 1-5 and 8-9.

FIG. 12A is a more detailed block diagram of a fifth portion of the system shown in FIG. 1, for implementing the integrated modes of operation shown in FIG. 11.

FIG. 12B is a table useful for understanding operation of the system portion shown in FIG. 12A.

FIG. 13A is a block diagram of a sixth portion of the system shown in FIG. 1.

FIG. 13B is a table useful for understanding operation of the system portion shown in FIG. 13A.

FIGS. 14A, 14B and 14C are flow diagrams useful for further understanding operation of an aspect of the system shown in FIG. 1.

FIG. 15 is a flow diagram useful for further understanding operation of another aspect of the system shown in FIG. 1.

FIGS. 16A and 16B are flow diagrams useful for further understanding operation of an aspect of the system shown in FIG. 1.

.FIG. 17 is a plan view of a control panel for the system shown in FIG. 1.

FIGS. 17A-17R are schematic representations of display screens generated in use of the system shown in FIG. 1.

Turning now to the drawings, more particularly to FIG. 1, there is shown a real-time disk system 10 for storing and playing back D1 digital 10-bit 4:2:2 component video and audio signals from magnetic storage disks 12. D1 is a shorthand notation for the RP-125 525 lines/frame digital video standard, and the compatible EBU 601/656 625 lines/frame standard. The system 10 has a main channel subsystem 14 with an associated smooth motion option 16 and a second or key channel option subsystem 18 with an associated smooth motion option 20. A control subsystem 22 is connected to the main channel subsystem 14 and the second channel subsystem 18. An audio option subsystem 24 is also connected to the control subsystem 22. A control panel 26 and associated floppy disk option 28 for the channel 14 are connected to the control subsystem 22. A second control panel and associated floppy disk option (not shown) for the channel 18 are also connected to the control subsystem 22.

Serial and parallel D1 digital video inputs 30 and 32 and outputs 34 and 36 are connected to each of the channels 14 and 18 and to the control su. bsystem 22. Monochrome analog input 38 and monochrome analog output 40 and 41 are also connected to the channels 14 and 18 and to the control subsystem 22. Bidirectional RS-422 port 42, Ethernet port 44 and SCSI port 46 are connected to the channels 14 and 18 and to the control subsystem 22. An audio input 48 and an audio output 50 are connected to the control subsystem 22 and to the audio option subsystem 24.

The audio option subsystem 24 includes two high quality audio tracks to provide the audio reference for video editing. The tracks have analog inputs and outputs and are stored digitally. The audio normally plays synchronously with the video but can be slipped (offset). The audio tracks are meant for editing reference, and no capability is provided for audio editing.

In use of the system 10, video is meant to be chopped up and reassembled during the editing process, while the audio must stay intact. The audio is therefore stored on its own, standard computer disk in the disks 12.

FIG. 2 shows details of the main channel subsystem 14 and the second or key channel subsystem 18. In the main channel 14, the serial and parallel D1 input 30 is connected through an input board 60 to a video processing board 62. The video board 62 is connected by a bidirectional, 11.times.2 wide bus 64 to disk arrays 66 and 68. Output 70 of the video board 62 is connected to optional smooth motion processing board 72, D1 output board 74 and digital to analog (D/A) conversion board 76. Outputs 34 and 40 are respectively provided by the boards 74 and 76.

In the second or key channel, the serial and parallel D1 input 32 is connected through an input board 78 to a video processing board 80. A composite key input 82 is also selectively connected to the video processing board 80 through an A/D converter board 84. The video board 80 is also connected by the bidirectional, 11.times.2 wide bus 64 to the disk arrays 66 and 68. Output 86 of the video board 80 is connected to optional smooth motion processing board 88, D1 output board 90 and digital to analog (D/A) conversion board 92. Outputs 36 and 41 are respectively provided by the boards 90 and 92.

Control or CPU board 22 is connected to control panel 26 for the main channel 14 and control panel 94 for the second channel 18. Control panel 26 is connected to floppy disk drive 28, and control panel 94 is connected to floppy disk drive 96.

Details of the control or CPU board 22 are provided in FIG. 3. A main CPU 100, implemented with a Motorola 68340 type microprocessor, is connected bidirectionally to a computer bus 102. System memory 104, comprising RAM, ROM and EEROM, is also connected bidirectionally to the bus 102. Time code overlay generators 106 and 108 for the main channel 14 and second channel 18 are connected to the bus 102, and provide a time-code character generation facility impressed over the analog monitor outputs.. A color space digital signal processor 110, implemented with a Motorola 56000 type DSP processor, is connected to the bus 102 bidirectionally through buffer 112. Main channel 14 is connected to the bus 102 bidirectionally through buffer 114, and second channel 18 is connected to the bus 102 bidirectionally through buffer 116. SCSI and Ethernet ports 46 and 44 are also connected to the bus 102. The control subsystem 22 is configured for variable speed operation in both the forward and reverse directions.

Details of the video board 80 are shown in FIG. 4. The video board 62 is essentially a duplicate of the video board 80, with omission of the input from the A/D converter board 84. The outputs from D1 input board 78 and A/D converter board 84 (FIG. 2) are supplied through a FIFO memory 140 selectively to framestores 142 and 144. The head swap and key multiplexer 146 connects either to input or output of framestores 142 and 144, and to the video board 62 and to a time base corrector 148. The time base corrector 148 is connected to the disks 66 and 68 through a channel encoder/decoder 150. A disk read/write control 152 is also connected to the encoder/decoder 150.

The computer bus 102 (see also FIG. 3) is connected through a CPU input/output circuit 154 and a disk control microprocessor 156 with an ESDI interface to the disks 66 and 68. The computer bus 102 is also connected to the framestores 142 and 144 through an EDAC (Error-Detection And Correction) block 158 for random access to the framestores.

Outputs 160 and 162 from the framestores 142 and 144 are selectively connected to a vertical interpolator 164. The vertical interpolator 164 is connected to a blanking circuit 166. Output 168 of the blanking circuit 166 is D1 video. The outputs 160 and 162 from the framestores 142 and 144 are also selectively connected to the multiplexer 146 to provide inputs to the disks 66 and 68. Outputs from the disks 66 and 68 are provided through the multiplexer 146 at 170 and through the framestores 142 and 144.

In order to store and retrieve digital video signal information in parallel, without requiring any serial to parallel or parallel to serial conversion, the disk systems 66 and 68 must meet certain requirements. A D1 video signal in its native form runs at 27 Mwords/s in both 525 lines/frame (U.S. standard) and 625 lines/frame (European standard), and each word comprises 8-10 bits depending on the application. When a D1 signal is time-compressed to eliminate unneeded horizontal and vertical blanking intervals (but not data-compressed), it runs at a rate of 21 Mwords/s. This rate allows sufficient vertical-interval information to record DVITC (Digital Vertical-Interval Time Code). A 10-bit time-compressed D1 signal thus runs at 26 MBytes/s, and the storage target of 30 seconds of D1 video consumes 784 MBytes. This points to using a standard 1-1.2 GByte disk drive to meet the storage capacity requirement, and modifying it to meet the bandwidth requirement.

The real-time disk system starts with a standard 1.2 GByte, 51/4" magnetic Winchester disk drive, with ESDI interface. The disk has 6 individual platters, which present 12 recording surfaces. In ordinary disk practice, 1 surface is dedicated to servo use and the remaining 11 surfaces are read or written one at a time. But for real-time disk use, the disk is modified to access all 11 data channels at once, each with its own preamplifier, equalization, and channel encoder/decoder. This increases the disk bandwidth by a factor of 11, which is sufficient to support the aggregate data rate of 21 Mwords/s.

Each disk consists of 1923 cylindrical tracks, hereafter referred to as cylinders, and each cylinder spans 11 recording surfaces to present an 11-bit parallel signal. A cylinder has both the capacity and bandwidth to store 1 TV field of 4:2:2 video, so 1923 cylinders yield slightly in excess of 30 seconds of video in 525 L/F. The disk spins at the precise rate of 1 revolution per TV field, which is 3600 RPM in 525 and 3000 RPM in 625. The time to hop from a cylinder to its nearest neighbor, known as seek time (about 3 ms), is just short enough to allow recording of real-time 4:2:2 video in a contiguous fashion, but too long to perform "random-access" by seeking to an arbitrary cylinder. Therefore to perform true random-access record & playback, at least 2 disks in tandem are required. A video caching technique to make use of multiple disks is described below. Because this seeking process takes time that could otherwise be spent recording data, the data rate is boosted from 21 Mwords/s to 25 Mwords/s to account for this dead time.

The 1.2 GByte drives come with a standard ESDI interface whose primary use in the real-time disk is control of seeking and spindle-lock. An ESDI interface supports a maximum of 7 disks in tandem. 7 disks give 31/2 minutes of record and play time as the upper limit of an expanded system, but this limit can be doubled through the disk reconfiguration technique. This technique allows 2 RT3 video boards in a single real-time disk system to share their disk dam through an auxiliary reconfiguration channel.

Each picture has a data header recorded with it, containing such information as time-code, 525/625, field count, etc. Each disk has in addition 2 cylinders devoted to overall "directory" information. This is termed the disk header below, as distinguished from the individual picture headers.

The 11-bit signal is recorded raw, not error-corrected. To provide some measure of error-protection, the 11th channel is used to "pinch-hit" for any of the 10 data channels with a hard error, on a cylinder-by-cylinder basis. The majority of cylinders do not require any channels skipped since the number of disk defects is low. The chances of 2 or more channels with hard errors is minuscule; should this situation be encountered, then the channel corresponding to the more significant bit is skipped in favor of the spare (11th ) surface. All disk cylinders are expected to be ultimately usable. A table governing the use of surface-skipping is placed at the disk header of each drive, determined when the disk is first formatted.

Pictures recorded in 525 may not be played back directly in 625, or vice-versa. This is because the same record clock is used in both standards, but since the disk rotational rate varies, the apparent played-back clock rate will be incorrect when the video standards are crossed.

Random Access and Disk Caching

A single disk does not allow true random-access playback of video, due to inordinate disk seek times that exceed the 3 ms allowed for continuous play. Therefore at least 2 disks are required to achieve random-access playback. There are two techniques available to record and play back video over multiple disks in such a fashion as to allow random-access playback of any imaginable sequence of video fields.

The first technique doles successive recorded fields onto each disk in succession in a "round-robin" manner. Thus with an even number of disks, random-access playback can be achieved as long as access takes place at a frame (field pair) boundary; 1 disk plays while another seeks. However, this technique can fail with an odd number of disks, for 2 successive fields may now reside on the same disk, but widely separated.

The second technique, known as disk caching, records a clip (sequence of fields) onto the first logical disk in the system for as long as possible, then records on the second disk for as long as possible, and so on. However, not all disk cylinders are given up for recording, rather a small number are reserved for caching. On playback, if 2 fields are required that reside on a common disk but cannot be fetched without excessive seek time, then before the playback is attempted caching is performed. Caching consists of writing an identical copy of the TV field which may not be reached in time onto another disk in its reserved cache area, so that during playback this second disk may provide the needed field without delay. A cached field may be required at each break in continuous play, and also at the end of a sequence if looping is to occur. Thus if a real-time disk user creates an edited video sequence consisting of N individual smaller sequences, then up to N cache fields may be required. A practical limit to this number N is 100, for edit segments are rarely composed of more than this many short segments. By reserving 100 cache field cylinders on each disk, out of 1923 cylinders at least 1800 are left for recording, and the 30 second record/playback goal is still met.

Disk caching has the disadvantage that, once an edit sequence is defined, a short time must elapse while the real-time disk automatically assembles its cache fields through hidden disk read/write operations. However, the caching technique holds two key advantages over the "round-robin" technique. First, caching works with any number of disks in a multiple-disk system, while round-robin works only with an even number. Second, caching records video sequences in contiguous fashion on disk while round-robin breaks them up immediately. If the disks are later reconfigured, for example a disk given up for key channel use, then large portions of an original recording are left intact and may still be played, while in round-robin all previous recordings are effectively lost.

The real-time disk basic unit can hold up to 3 disks in a disk tray, due to physical constraints. The disks can be allocated in any combination between main channel and second channel, by plugging cables appropriately, but the common configurations are listed below:

To achieve more storage, extension disk trays are added with extension cables. The ESDI guidelines must be followed; no more than 7 disks total on either the main or second channel. Record/play time in a 2-channel system can be increased for either channel when the other channel is idle, by accessing the idle channel's disks through the disk reconfiguration channel. Record/play time can be further increased by connecting 2 or more units in tandem, with the video output of the first unit feeding the video input of the second unit, and so on. Control of the aggregate joined units is assumed through a single control panel, and the RS-422 control panel communication is chained from unit to unit.

The second channel when used as a key channel is assumed 4:0:0 (monochrome), and so recording will take place somewhat differently from the main 4:2:2 channel so as not to waste disk space. The key channel has its own data bus and disk system, separate from the main channel. Since the bandwidth of the key channel is 1/2 that of the main channel, only 5 bits at a time are recorded in either the 5 more-significant or 5 less-significant bit positions, for the entire cylinder. This multiplexing technique permits recording a second, independent TV field in the remaining 5 bit positions at a later time without disturbing the first recorded field. By recording field 1 of a frame in the upper 5 bit positions, and field 2 in the lower bit positions, a cylinder may hold an entire frame (field pair) of video. During playback, all 10 bits are played back. This effectively captures 1 frame during the time that only 1 field is needed. With this technique, random-access may be achieved with only 1 disk, for the second field may be displayed during the time the disk is seeking a distant cylinder.

Multiuser Capability

By utilizing the second channel to its full 4:2:2 capability, the disk system may support 2 independent users each with their own control panel and set of disks. This permits 3 modes of operation:

a) both users may operate completely independently, constrained only that both must work within the same video standard.

b) when one user is absent the other may have access to the entire disk system disk capacity, thus boosting his record/play time. The system is configured through the control system to let the sole user employ the disks of the absent user through the reconfiguration channel.

c) when one user is absent the other may have access to the second channel as an independent record/play channel. This lets one channel record while the other channel is playing, for example, and thus allow multiple-generation image compositing by interposing an external compositing module between the played channel output and the simultaneous recorded channel input. After each pass the record/play role of each channel is swapped, and successive generations of compositing may be built up without loss of signal quality.

CPU board 22 controls the operation of both users by multitasking the control software.

Record Path

Referring to FIG. 6, Record Path. D1 4:2:2 video at 10 bit per pixel precision enters Input FIFO (first-in, first-out memory) 600 at the left. This FIFO affords a nominal 1 TV line delay .+-.1/2 TV line, to allow for potential input video mistiming. Video then enters one of 2 Framestores 604 or 606, which are used in a "ping-pong" manner to store alternate video frames. These framestores are constructed from 256K.times.4 VRAMs which permit either a single-bandwidth sequential video input or video output, as well as simultaneous random access through the CPU port. The random-access port of each framestore connects to the local computer bus through EDAC (error-detection-and-correction) block 608, and then to the main CPU bus through CPU interface block 602. Video from the framestores is output for viewing through vertical interpolator 630 and blanking circuit 632, shown in Playback Path (FIG. 6); no vertical interpolation is required while viewing this "live" video. The same framestore video is passe. d to disk Head-Swap and Key-Channel Multiplexing circuit 610. The function of this block is to repeat a disk data channel that might be unreliable onto the spare 11th data channel, to enhance overall data integrity, and also to cross-connect to another similar board through the Crossover channel to make use of that board's attached disks. Video of 11 bits per pixel emanates from 610, and passes to TBC (time-base correction) FIFO 612. This FIFO directly matches pixels at video rate (27 MHz) to the disk-channel data rate (25 MHz), and is 1/2 TV field .+-.1/2 field in depth. It consists of a single 1 field.times.4 bit FRAM (field random-access memory, part TI TMS4C1050) per pixel bit plane, 11 bit planes in all. The luminance (Y) and chrominance (C) signals from each bit plane then pass to Data Write Control PAL 614, which interleaves the Y and C into a single bit stream for recording, and also generates timing signals for Channel Encoder/Decoder 616. The Channel Encoder/Decoder is used as an encoder while recording, and is part SSI 32D5372 used one per bit plane. The Channel Encoder generates a (1,7) RLL (run-length-limited) channel code, which is an industry standard for magnetic disk recording. Since the (1,7) code generates 3 channel bits for every pair of dam bits, the Channel Encoder is fed with a 3.times. data clock of rate 75 MHz from Master Oscillator 618. The Disk Control microprocessor 620 generates the disk ESDI control signals. Both control signals and data signals are combined in a single cable to connect to the disk subsystems. Up to 7 disk subsystems may be attached, due to the ESDI control limit of 7 devices. Of the 7 possible disk subsystems, only 1 will be recording or playing back at any instant; the remainder will either be idle or seeking a distant cylinder. In the disk subsystem, the disk is modified to bring all 11 data head connections out, and each Disk Head 624 is supplied with its own Record and Play Amplifier 622, part SSI 32R4610.

High-level control for recording and playing is assumed by the system CPU 100 which controls the CPU bus. The Disk Control microprocessor 620 takes its instructions from the CPU in high-level description such as which disk subsystem to enable, which disk cylinder to seek, read vs. write, and so forth, and in turn both controls the ESDI bus at critical timing points so as not to overtax the system CPU and synchronizes the disk rotation rate to the video field rate. Each recorded cylinder is prefixed with identifying header information, which is written into Framestore 604 or 606 directly ahead of the video through the random-access port with EDAC 608 turned "on." The EDAC imparts a high degree of data confidence to the header information. The EDAC method consists of simply repeating each bit 5 times, then taking a majority-vote among the 5 received bits upon playback to decode. This EDAC method of repetition is very powerful, but seldom used in other engineering practice due to excessive redundancy. However in this application the redundant information is insignificant in relation to the sheer mass of video information contained within a single disk cylinder.

Playback Path

Referring to FIG. 6, Playback Path. Since the real-time disk system does not allow simultaneous record and playback of video, the same elements used in the record path are reconnected backward during playback, thus reducing system cost. Each of 11 Disk Heads 622 connects to Record and Play Amplifier 622, and the analog dam then sent to Equalizer and Detection stage 626, part SSI 32P541. Data detection is performed in a conventional way, and need not be described here. Disk Control microprocessor 620 plays the same role as it does during recording. Detected digital data is sent to Channel Encoder and Decoder 616, which locks a PLL (Phase-Locked Loop) to each data channel, and decodes the (1,7) RLL channel code. Data Read Control PAL 628 deinterleaves each data stream to recover Y and C separately, and passes Y and C to TBC FIFO 612 for time-base correction. Data out of the TBC FIFO, now at video rate, passes to Head-Swap and Key-Channel Multiplexing circuit 610 which substitutes the 11th data channel for any individual data channel previously deemed unreliable. Video data is now at 10 bits per pixel, and enters one of 2 Framestores 604 or 606, which are used again iri a "ping-pong" manner to store alternate video frames. The cylinder header information is recovered by reading the framestore holding the appropriate cylinder while turning "on" EDAC block 608, and reading the header information from the system CPU (not shown) through CPU Interface 602. Video from the framestores passes to Vertical Interpolator 630, whose function is to shift a field's video up or down 1/2 line to avoid the picture hopping experienced when a TV field is displayed during the opposite field time, i.e., an original TV field 1 displayed during TV field 2. This situation is termed opposite interlace polarity, and might arise, for example, when performing fast-motion playback by simply dropping TV fields. When no opposite interlace polarity is encountered, the Vertical Interpolator merely passes the video unaltered. From here the video passes through Blanking circuit 632, which performs horizontal and vertical blanking in accordance with the D1 standard, and then to the output.

Motion-Adaptive Smooth-Motion Processing Option

In this application, the SMO-MO Option is represented by devices 16 and 20 in FIG. 1. For the case of smooth motion portrayal, the new proposed system takes the concept of motion-adaptive standards converters a step forward by creating not one, but many new fields/frames in between two originals. The two original frames are not necessarily contiguous in time, but for most cases they will be since this is the way frames are stored in the real time disk.

The motion information is derived from localized absolute frame differences. Processing of the motion field starts by rectifying the frame differences and performing a two-dimensional filter in order to eliminate higher order aliasing frequencies that result naturally from rectification and rounding. This two-dimensional processing serves to transform the difference signal into a valid representation of motion.

The motion field signal serves as input to a non-linear transfer function which makes a dynamic soft-switch between two different processing modes. Such soft-switching characteristic is controlled by threshold parameters that can be set dynamically via external control.

For areas of the scene where strong motion is detected, new frames (in between) are created by repeating the previous or current original frame, with the aim of avoiding severe blur that would be caused by temporal interpolation. On the other hand, for areas where little motion is detected, the new frames are created via proportional temporal interpolation. It is assumed that there is little difference in these areas. From one frame to another, there would be virtually no loss of resolution or blurring caused by this processing. Temporal interpolation has the added advantage of providing noise reduction. The majority of cases will fall somewhere in between these two extremes; here, a combination of temporal interpolation and frame repetition will be done according to the value dictated by the non-linear transfer function. The net effect of this processing shall be to `squeeze` the time display closer to the original frames and therefore providing quick but smooth transitions in areas of relative large motion, without the inconvenience ofjudder.

Devices 16 and 20 in FIG. 1 can be configured dynamically to perform motion-adaptive recursive noise reduction. The same motion detection signal path that produces the motion field can be used to implement recurslye noise reduction when the Smooth Motion option is not being used. Alternatively, two different devices can provide both options by simple cascading of output of one board to the input of another.

The SMO-MO processing board is able to perform film-to-video transfers as described below. In order to improve the judder created by the traditional 3:2 pull-down conversion, the new system performs linear proportional interpolation along the time axis. In this case the same hardware acts in this special mode for material that is known to have proceeded from a Telecine Machine (already converted from film to video). The process involves `undoing` the 3:2 pull-down (or any other pulldown sequence) and temporally interpolating the new frames according to the proper position dictated by sampling-rate conversion theory.

In addition to the features mentioned above, the same piece of hardware can be configured to provide a temporal averaging over N video frames. The output rate of frames will be reduced. by N, and therefore will produce an effective speed up unless the results are stored back to the real-time disk storage or routed to another similar hardware for display at any desired speed factor.

Explanation of the SMO-MO Hardware

The hardware for Smooth Motion Processing with added features mentioned above is able to support several operating modes: motion adaptive time interpolation, motion adaptive noise reduction, motion adaptive film-to-video transfers, and N-frame temporal average.

Motion Adaptive interpolation:

This mode of operation is used to provide the capabilities of slow motion and fast motion rendition. In the case of slow motion portrayal of information stored in disk, it creates the necessary in-between frames corresponding to a slow-down factor determined by the user. For instance, if the user specifies a slow-down factor of 10, then 9 new frames will be created in-between any two actual frames in the specified sequence. It is possible to specify non-integer factors, if desired. The effect of the motion adaptive soft-switch or fade between temporal interpolation and frame repetition is controlled by threshold levels set externally.

The same hardware is able to perform recursive noise reduction based on any two input fields. When not operating in smooth motion processing mode, the user of the system gets the option of continuous noise reduction based on the processed motion field. The advantages of non-linear fading are thus realized for the conventional recursive noise reducer implementation.

It is also possible to perform motion adaptive noise reduction on film material from a Telecine by undoing the effect of the 3:2 pulldown (br any other pulldown sequence), carrying out noise reduction, and reforming the 3:2 pulldown sequence. As explained below, the film material can be adaptively temporally interpolated with two identical hardware devices connected in series and set to the proper operating modes.

As mentioned above the simplest way of performing these film transfers is to use the ubiquitous 3:2 PullDown where conversion is made from film rates of 24 or 25 frames per second (Fps) by simply resampling by repetition to a higher temporal frequency and displaying the information at that new temporal frequency. A better way is to perform electronic time interpolation to create frames that are correctly `located` in time according to the 2/5 and 5/2 time relationship between film and video temporal sampling frequencies, in the 525 L/F standard. This is simply a special case for the Smooth Motion processing option, and can be handled easily. Input video is reconstituted to the original frame rate, then the intermediate frames are generated by motion adaptive temporal interpolation as explained above.

N-Frame Time Averaging

Under certain circumstances, and in order to both perform either N-frame noise reduction or artistic processing of video sequences, the hardware is able to perform arithmetic averaging over any set of frames N. There will be a delay of N input frames between each output frame, accompanied with a speed up by N. Processing options with more than one hardware device:

A combination of similar hardware will allow the realization of several medes of operation that are not possible with a single device. For instance, with only two identical hardware units it is possible to accomplish the following features:

For video sequences recorded from conventional Telecine machines: reconstitution to film rates (24 Fps) without degradation, motion adaptive noise reduction, and motion adaptive transfer to video rates.

For video sequences recorded at normal video rates: N-frame averaging and speed up to normal play; Motion adaptive noise reduction and smooth motion slow down or speed up.

After an introduction to the basic concepts used in the SMO-MO option (16 and 20, FIG. 1 ), a more detailed explanation of the hardware implementation (72 and 88, FIG. 2) shall follow.

Motion-Adaptive Temporal-Linear Interpolation

Temporal-linear interpolation in this application refers to the process of creating new video frames from linear combinations of other video frames. The originating video frames are typically, but not necessarily, contiguous in time. These frames will be referred to as `original` or `base` frames in this application.

FIG. 7 illustrates the case of linear interpolation between two frames A and B to produce a plurality of new frames. Frame B occurs first in time, and frame A occurs T.sub.frame seconds later (Tframe may be 1/30 sec., i.e., the frame time for NTSC). The newly created frames are linear first order combinations of the base frames.

In the example illustrated in FIG. 7, the newly created frames will be displayed at the normal rate, therefore, they will be separated in time by T.sub.frame seconds. It should be clear that the appearance of slow motion will be generated since the original frames will be displayed at a rate of 1/(N*T.sub.frame). In this example, the parameter N is the slow-down factor; and the parameter n indicates the new frame being temporally interpolated.

It can be said that frame B smoothly fades-over frame A. This fade-over or proportional interpolation in the time domain is used for cases where there is no significant motion on a pixel-by-pixel basis from frame B to frame A. In areas where there is strong motion, Frame A or B is repeated depending which one is `closer` in time to the newly generated frame n. Therefore, there are two reciprocal fade-overs: the first, done between A and B so that the newly created frame I is

where

and the second, between the interpolated result I and the frame repetition F, which has the value of A or B depending on whether frame I belongs to the interval 0<T<0.5 or 0.5<=T<1.0. Variable M denotes the amount of motion detected between frames A and B:

where

Variables A and B represent the values of two pixels (PA and PB) at the same spatial position, but delayed by T.sub.frame.

The formula above corroborates the fact that when there is no motion detected (M=0), the output pixel P will be simply the linearly interpolated result I. When there is strong motion detected between A and B (M=1), the value of F will be used as output. For all other cases in between M=1 and M=0, a reciprocal combination of the two signals I and F will be used.

A non-linear transfer function which can be easily implemented as a dynamic look-up table guarantees that the degree to which M is effective can be tailored to specific situations in which a system operator may require to artificially set the value of M to 0 or 1, or at various values in the interval.

The process described in the equations above is illustrated in the diagram and formulas of FIG. 8. However, the block diagram of FIG. 8 is not the most economical implementation for a hardware system.

The formulas indicated at each stage of the signal path in FIG. 8 represent the signal processing done on picture data. The output OUT on a pixel-by-pixel basis can be represented by the following equation:

By substituting the equations for I and F:

This equation simplifies to:

Where, for the interval 0<=n<N/2:

And, for the interval N/2 <=n <N:

As seen from the previous equations, the values of F and T depend on whether intermediate frames are created for the interval 0<n<N/2, or the interval N/2<=n<N.

The block diagram of equation (1) is represented in FIG. 9A, and forms the basis for the hardware implementation. In this figure, digitized video is delayed by one frame-time Tframe by means of device 404. Device 408 is used to multiplex frames A or B; its output corresponds to variable F in the equations above. Video data is subtracted by device 406 and such difference signal is used to detect motion by means of devices 414, 416 and 418, which perfoms rectification, two-dimensional filtering and non-linear transfer function (NLTF), respectively. The output of device 418 constitutes a representation of the amount of motion that occurs between two frames, and is the signal which dictates the adaptive temporal-linear interpolation process. The difference signal out of 406 is mixed with the signal out of 418 by means of 422, which in its simplest form can be a digital multiplier/accumulator (MAC). Device 424 performs the addition of signals from 408 and 422, and corresponds directly to the value Pout represented in equation (1).

The block diagram of FIG. 9A can be directly implemented in hardware, but as explained below, a number of useful features can be added to such system. In such case, the individual processing blocks are more complex since they must perform various signal processing operations depending on the operating mode. The processing extensions of the adaptive technique described above are referred to as `Smooth Motion Processor with Features`, which is implemented as a single processing board integrated into the real-time disk main system described above.

Motion-Adaptive Film-to-Video Transfers The most common way of converting motion picture film material to television video frames is done by 3:2 pull-down, as shown in FIG. 14a and 14b. Converted television frames TVa, TVb and TVe are generated from single film frames, therefore, there is no interfield motion and thus no motion artifacts between odd and even fields. On the other hand, converted television frames TVc and TVd are known as `jittery frames` because the odd and even fields are created from different film frames, and therefore, there is the possibility of interfield motion shown as a jitter effect. In the 3:2 pull;down technique, three television fields are created from even-numbered film frames; and two television fields are created from odd-numbered film frames. Besides jitter effects, this conversion causes judder (abrupt motion artifacts when large portions of the scene move from one frame to the next) for scenes that originally depicted smooth motion on film media.

The Smooth Motion system proposed can be used in the SMOoMO mode to reduce the jitter effects indicated above. This is shown in FIG. 14c, where new TV fields to replace the `jittery` ones are created via motion-adaptive linear interpolation between the previous and following fields. FIG. 14c shows that Odd field 5 is created from previous Odd field 3 (originated from film frame 2) and subsequent Odd field 7 (originated from film frame 3). Even field 8 is generated in similar manner. FIG. 15 shows this same process on a field basis.

Although the technique depicted in FIGS. 14A-14C and 15 will show improvements in jitter for two frames out of five, smooth motion scenes will still show judder due to the repetition of entire film frames at the incorrect temporal positions for the television field rate. The Smooth Motion processing system can be used to `position` the newly created television fields in the `correct` temporal position for the smooth portrayal of motion from the originating film frames. This process is shown in FIG. 16B (FIG. 16A is simply a reiteration of the first technique outlined in the above paragraph). Each new field is generated proportionally with the factors indicated by the arrows. This proportional interpolation is motion-compensated with the previous and subsequent frames and, in the limiting case of M=1 (large motion detected between frames), it reverts automatically to the familiar 3:2 pull-down case (simple repetition of entire frames).

The two techniques mentioned above assume the correct identification of video fields that have been generated with the 3:2 pull-down technique. The Real-Time Disk System must `undo` the 3:2 pull-down and present the `original` frames to the Smooth Processing System. Correct identification is done by noting two contiguous `jitter frames`. The next `stable frame` will be `Frame 4` as indicated in FIGS. 14A-14C. The correct identification can be done visually by a system operator, or automatically by using information from the Global Motion Processor 420 indicated in FIG. 12A. The automatic identification of the b 3:2 pull-down sequence is not guaranteed to work with 100% success, but for most material it will correctly recognize the `jittery frames` and therefore the correct sequence.

It has been thus shown that the Smooth Motion Processing system can be used to accomplish motion-adaptive film-to-video conversion and that this process will produce better results for special cases where the `traditional` 3:2 pull-down method shows jitter and/or judder artifacts.

Hardware Description for Smooth-Motion Processing Board with Features

This processing board is referred to as RT4 SMO-MO, devices 72 and 88 in FIG. 2; its processing blocks are depicted in FIG. 9B. This board can operate in various modes described in subsequent sections and depicted in FIGS. 10A and 11. The operating modes are Smooth Motion Mode (SMO-MO), Recursive Noise Reduction Mode, and Average/Integration Mode.

FIG. 11 depicts the concept of integrating the above operating modes into a single system.

Each operating mode is selected by CPU I/O device 429, which receives instructions from device 22, RT2 CPU Board. Although the digital input signals are 10-bit wide, this board processes the information with an accuracy of 12-bits internally.

The correspondence of various signals in the system with variables in the formulas described above is as follows: (note that the value of T is input into device 8 by means of device 11 ):

The signal processing flow indicated in FIG. 9B starts with the input signal 400 and delayed signal 430 being processed by Arithmetic Processor 405, which yields a weighted sum or difference according to the operating mode. The output of 405 is used by 413 to create a motion detection signal. Device 408 is used to select the value of F as discussed above. Devices 410 and 412 serve to delay signals 433 and 435 so that they correspond to the same processing pixel at the input of device 421. Signal 432 out of the motion processor is used by 421 to perform adaptive temporal-linear interpolation. The value of T, as well as other parameters in the system, including the selection of various NLTFs is effected by means of 429. Signal 441 can be used as feedback to the input stage. When multiplexer 402 selects signal 441, a feedback loop is formed. This feedback loop is useful to perform recursive noise reduction. Signal 44 1 is also rounded to 10-bits and blanked if necessary for proper display by means of device 426. At the output stage, device 428 provides a constant signal delay from input 400 to output 438.

A more detailed block diagram of the hardware implementation is depicted in FIG. 12A. Device 22 (FIG. 1) controls the values of the system parameters, the coefficients K1 through K4, the settings of the multiplexer/switches SW1 and SW2, and the output of the non-linear transfer function NLTF. The table of FIG. 12B shows the particular settings for each operating mode. The system is designed so that the settings in the table need to be modified no more often than on a TV field-time basis.

The various operating modes are described below. Refer to FIGS. 12A and 12B.

Smooth-Motion (SMO-MO) Mode:

The setting of SW1 to position C, and the values for K1=1 and K2=-1 effectively cause the input signal 400 to be subtracted from signal 430 (which is the input delayed a whole frame by device 404). The value of K3=1 permits this difference signal to be processed by devices 414, 416 and 418 for properly estimating motion between signals 400 and 430.

Device 414 is a rectifier which generates the absolute value of the difference signal. This signal is low-pass filtered by device 416 in the horizontal and vertical direction in order to eliminate high-frequency alias components produced by rectification. Device 416 also transforms the signal into a close representation of motion, i.e., a motion field between signals 400 and 430. The non-linear transfer function (device 418) makes the decision as to what constitutes large or small motion between frames, and contains multiple threshold parameters (selected by 429, FIG. 9B), which indicate the critical transition regions from no-motion to full-motion. A soft, non-linear transition is performed in each region indicated by the threshold parameters. The output of the non-linear transfer function constitutes the motion field signal 432.

The value of K4 is (n/N) or (-n/N) depending on the number n of the frame being created in the interval [0:N/2] or [N/2:N]. The value of K4 increases for the first interval, and decreases for the second, as indicated in a previous section. The motion signal 432 can also be modified externally at input 434. Signal 436 is the motion compensated linear interpolated coefficient. This signal is also used by other channels (Chroma and Key) for smooth-motion processing and noise reduction (explained below).

Global motion processor 420 is used to calculate the total sum of the motion field between frames A and B, as well as the minimum pixel motion detected. The minimum and total values are used to change system characteristics during scene changes and to change the NLTF according to scene content. There are other uses derived from the global motion processor, which are explained in other sections below.

Switch SW2 is changed from B to A every N/2 created frames, and added to the output of device 422 by device 424. This produces the desired motion-compensated fade-in between linear interpolated and original frames, as the case may be on a pixel-by-pixel basis.

General purpose delay elements 410 and 412 are used for properly aligning the signals in time. Rounding to 10-bits is performed by device 436. Variable frame delay device 438 is used to provide a constant delay between input 400 to output 438. The maximum delay between input and output for this operating mode is 2 frame times.

Digital Nois.e Reduction (DNR) Mode

All the switches and coefficients used for the SMO-MO case are used here, with small differences indicated in the control table of FIG. 12B. There are four main differences between the DNR and SMO-MO modes. First, SW1 is set to position D, which means that the signal out of device 424 is fed-back to the input of frame delay device 404; therefore, the output of device 406 is a first-order IIR (Infinite Impulse Response) filter. Second, SW2 is always set to A, which is a requirement dictated by the concept of recursive noise reduction as illustrated in FIG. 10B. Third, the non-linear transfer function parameters are different and are influenced in different ways by the global motion processor 420. Fourth, the total delay between input 400 and output 438 is one frame, due to the fact that there is no need to wait for processing two frames in order to create the desired noise reduction for the current frame.

Average (AVG) and N-frame Integration (INT) Modes:

The purpose of averaging over N frames is to produce special effects and noise reduction for video clips where there is no motion, although there is nothing that prevents the user to utilize the system in this mode for all types of inputs. Noise reduction in this mode is possible because it has been proven that the effect of random noise can be `averaged out` when the value of N is very large. Integration is used when dealing with material obtained at low illumination and for which it is desired to bring the average signal level to a higher value.

Control settings for both modes of operation are essentially the same, with the exception of K1. The coefficient K1 performs the averaging operation on a sample-per-sample basis in the AVG mode.

The value of K3 is set to zero since there is no need to perform motion detection. This value also selects the output of the NLTF as indicated in the control table of FIG. 12B. The value of K4 can be set to 1 as indicated, but it can be changed just as easily to provide a constant offset when needed.

The total delay between input and output is dependent on the number of frames N being averaged or integrated.

Transparent (TRANSP) mode:

When this system is not in use, i.e., none of its features are desired, the input signal 400 is passed through the hardware to the output 438 with a delay of exactly one frame time.

Chrominance and Key Channels:

The processing for the chroma and key signals is depicted in the block diagram of FIG. 13A. The motion detection path has been eliminated. The motion information is derived from the luminance channel and is input through EXT MOTION and optionally modified by K4, if desired. The control table is shown in FIG. 13B, and is similar to the control table for the luminance signal described above.

Control Panel

FIG. 17 shows further details of the control panel 26. FIGS. 17A-17R show display screens generated in use of the system 10. The RTD control panel 26 consists of a 42.times.8 character display 200; a rotary control 202 to support jog, shuttle and variable-speed functions; 5 "soft" keys 204, whose meaning depends upon individual menu context, lying below the character display 200; 15 keys 206 in the keypad group, without indicator LEDs; and 40 keys 208 in the keyboard group with indicator LEDs, logically grouped into 5 smaller groups: the playback mode group, the transport control group, the segment group, the setup group, and the remainder.

Keypad Keys

Clear - clears any keypad entry.

Enter - completes the numerical entry for the function selected.

key - used to define decimal values, as in variable speed play.

key - used to delineate time-code numeric fields.

.+-.key - pressed once to set negative numbers, twice to set positive numbers for moving in field/frames increments when used before GoTo. ".+-.", (number), then "GoTo" will move the disk back (-) the number of frames entered; ".+-.", ".+-.", (number), then "GoTo" will move the disk ahead (+) the number of frames entered.

0-9 keys - to enter numerical values.

Keyboard Keys

PLAYBACK MODE GROUP

Normal Play (LED on/off) - when ON allows playback access to the entire disk; also used to return from sub-menus to the normal play (default) menu.

Clip Play (LED on/off) - when ON limits the disk playback to the current clip. GoTo can be used to move to another clip by clip # or by time-code/disk time-line. A clip is defined automatically as any recording made at one time with no changes in the record setup. Clips can be trimmed in the Clip Play menu.

Segment Play (LED on/off) - when ON enters the SEGS Play menu and limits the playback to disk tracks as defined in the current segment list.

Cine Play (LED on/off) - when ON, and when a clip has been identified as 24 or 30 fps film, and when the master frame of a 24 fps clip has been marked, several special playback modes will be used for real-time and especially non-real-time playback and jog of the clip. Other advanced modes will be possible with the "Smooth Motion" (SMO-MO) option.

When Cine Play is selected and the current clip has not been previously "marked", it should bring up a sub-menu which provides for identifying a clip as 24 or 30 fps film (or animation); if 24 fps it will demand the identification of any "master" frame of the 3:2 TV/film frame sequence.

If SMO-MO is installed and turned on, real-time playback of 24 fps film can introduce new "mixed" fields in place of the duplicate 3rd field of the 3:2 sequence. If SMO-Mo is installed and turned on, playback of 30 fps film OR animation can introduce "smoothed" fields in real-time and non-real time playbacks.

NOTE on Playback Modes: Normal, Clip & Segment are mutually exclusive modes. Cine Play can be used with the other playback modes.

Play.fwdarw.(LED) - plays the disk video at 1X speed, unless Vari-Speed is turned ON.

Play.rarw.(LED) - plays the disk video at 1X speed, unless Vari-Speed is tumed ON.

Stop (LED) - stops any play or record operation in progress. If the disk has been under external editor or EtherNet control, returns control of the disk to the CP. None of the transport control LEDs should be lit while under external control.

Step.fwdarw.-will step the disk forward along its time line either one field or one frame as determined in the Output Setup.

Step.rarw.will step the disk backward along its time-line either one field or one frame as determined in the Output Setup.

Loop (LED on/off) - when ON will allow all plays, steps, etc. to reach the end of the disk, clip or segment and automatically jump back to the first frame, in effect creating a continuous loop of video.

Ping Pong (LED on/off) - will allow all normal or variable speed plays to automatically reverse when reaching the end of the disk, clip or segment, and reverse again when back at the beginning.

NOTE: "Loop" and "PingPong" are mutually exclusive. If one is selected the other is automatically de-selected.

Vari-Speed (LED on/off) - when ON, all plays will be at the speed entered by the keypad. This entered value will only be changed by entering another value (#'s followed by pressing "Vari-Speed) OR by using the position ring of the transport knob to increase or decrease the entered value while in play. Vari-Speed can be turned on while the disk is playing, which causes the speed to go from the 1X normal to the speed set.

SMO-MO (LED on/off) - enables the operation of the optional Smooth Motion board if installed. Brings up the Smooth Motion setup menu.

Record (LED on/off) - followed by "Play.fwdarw." will begin recording the number of frames as entered in the Record Length. Pressing "Stop" while recording will end the recording. Pressing any other button before "Play.fwdarw." will de-select "Record".

Record Lock (LED) - will be used in conjunction with the "MARK" keys to protect tracks of the disk so they cannot be erased. The LED will turn ON any time a protected section of the disk is or would be accessed by any operation.

GoTo (LED) - when disk is stopped will send disk to the frame/field as entered by the keypad numbers. The sequence can be either "GoTo", "(numerical entry)", then"ENTER", OR "(numerical entry)" and "GoTo".

In Segment Play; "GoTo", "#", "ENTER" will send the disk to the first frame of the segment number entered.

".+-." used ahead of the numerical entry will increment the disk by the time value entered.

"Field" used after the numerical entry will increment the disk by the number of fields entered.

Shuttle (LED on/off) - when the LED is OFF, the rotary control is used for jogging (position control). When the LED is ON, the rotary control is used for shuttle (speed control).

Mark In - Marks the current disk track as the first field/frame of a loop or segment. Can also be used to mark a record in-point. Would suggest that numerical keypad entry of a time-line value followed by "Mark In" would work the same without the disk needing to physically go to that frame.

Mark Out - same as "Mark In", but for the last frame of a loop or segment.

Seg Insert - hitting Seg Insert will identify the current "Mark In / Mark Out" points as the start and end of a new segment, and insert that new segment into the segment list. If in Normal Play mode the new segment will be added to the end of the segment list. If in SEGS Play, the new segment will be inserted ahead of the currently highlighted segment.

Seg Edit - Brings up the Segments Editing menu (delete, copy, seg/speed, move).

Insert Clip - will take the current clip and insert it as a new segment in the segment list, without needed to manually mark the in- and out-points.

Bypass (LED on/off) - toggles the output video between the disk output as determined in the OUTPUT Setup and the input as determined in INPUT Setup.

GRAB (LED on/off) - if pressed while playing disk video will "freeze" the output video. If pressed while in Bypass, will freeze the input video. The video will stay frozen until GRAB is pressed again.

Record Setup (LED) - brings up the menu used to set up all record enables, to allow recording any combination of video, time code and audio.

Input Setup (LED) - brings up the menu used to set up and mark the input source to be recorded.

Audio Setup (LED) - brings up the Audio setup menu; lit anytime internal or external audio is synced/locked to disk playback.

TC Setup (LED) - brings up the Time Code setup menu; use VITC, LTC or RTD time-line; adjust/slide timelines, etc.

Output Setup (LED) - brings up the Output Setup menu; used to set output mode field-frame-autoframe, field interpolation on/off, set output timing, and select 8- or 10-bit output.

GPI Setup (LED) - brings up the GPI menu; used to assign functions to GPI ins - Record, Play, Stop, Step.fwdarw., Step.rarw., macro #s.

Remote Setup (LED) - brings up the Remote Setup menu. Used to enable/set up all RS422 ports, set editor protocols, etc. LED will be on if the system is being controlled by any external device.

Dub/Dump (LED) - will bring up a menu which allows the control of an external device (VTR), and 1) marking the clip of the external device to be recorded, and executing the recording from the external device onto the RTD, or 2) marking the recording start point on the external device, and executing the transfer of material from the RTD to the external device. This must be frame accurate.

Backup (LED) - will bring up the menu which controls the setups for the SCSI ports, and the backup and restore operations of the SCSI device; also will be used to set Ethernet address, etc.

Diag/Test (LED) - will bring up the menu which contains all diagnostics and test patterns and routines.

Browse (LED) - in Normal, Clip or Cine Play mode will display clip keyframes. In Seg Play will display the segment keyframes. This button also brings up a menu which will allow other `browse` choices, such as last frames, next frames, bracket the current frame, etc.

Macros (LED) - brings up the Macro menu for recording and running macros, along with the macro edit sub-menu.

Attached hereto and forming a part hereof is an appendix, consisting of a source code listing in the "C" programming language of a system control and user interface program for the system 10.

It should now be readily apparent to those skilled in the art that a novel realtime disk system capable of achieving the stated objects of the invention has been provided. In particular,

1) No other disk, PTD (Parallel-Transfer Disk) or otherwise, has both the bandwidth and capacity to store 30 seconds of 10-bit D 1 4:2:2 video in real time.

2) The disk contains 11 data channels, used as 10 channels plus an extra error-protection channel. The video data is also 10-bit precision. This forms a direct match between the video data and the disk data without the use of complicated parallel-to-serial and serial-to-parallel data conversion, as used in the past.

3) Each of the 11 identical disk data channels is built using industry-standard magnetic-disk ICs, which keeps the cost low.

4) Internal control of the system is performed at two levels. The system CPU performs all user interface and high-level functions, while a dedicated disk-processor CPU manages low-level disk functions. This separation of duties greatly facilitates the real-time control implementation.

5) 2-channel operation within a single unit provides the following possibilities: 1 full 4:2:2 channel plus 1 simultaneous 4:0:0 key channel; 2 independent 4:2:2 channels which may record and play at once, to accommodate multiple-generation image compositing; 2 independent users each with their own control panel and disk system; or 1 user may take over the disks of the second user when that user is absent, and thus increase his record time. This provides flexible and cost-effective operation.

6) Record/play time may be increased in 3 ways: more disks may be added up to the bus limit (7 maximum); a 2-channel system can allocate all disks to 1 channel; and multiple units may be chained to provide effectively a single unit with the combined record/play time.

7) A single board can provide alternately real-time smooth-motion processing, noise reduction, frame averaging, or frame integration.

This new apparatus provides multiple improved features as indicated below, in a single piece of hardware.

It should further be apparent to those skilled in the art that various changes in form and details of the invention as shown and described may be made. It is intended that such changes be included within the spirit and scope of the claims appended hereto. ##SPC1##