Integrated Analog-to-Digital Converter

To these and to such other objects which may hereinafter appear, the present invention relates to an integrated analog-to-digital converter, as set forth in detail in the following specification and as recited in the annexed claims, taken together with the accompanying drawings, wherein like numerals refer to like parts, and in which:

FIG. 1 is a block diagram of the integrated analog-to-digital converter of the present invention;

FIG. 2 is a block diagram of the N-bit feedback coder of the present invention;

FIG. 3 is a schematic digram of the CMOS push/pull amplifier used as a gain element in the feedback coder of the present invention;

FIG. 4 is a block diagram of a preferred embodiment of the single bit feedback coder of the present invention;

FIG. 5 is a block diagram of a preferred embodiment of the second order single bit feedback coder of the present invention;

FIG. 6 is a schematic diagram of the delta sigma modulator of the present invention;

FIG. 6A is a schematic diagram of the static D-type flip-flop circuit employed in the single bit feedback coder illustrated in FIG. 6;

FIG. 7 is a block diagram of a generalized decimator including finite impulse response filter and sub-sampler;

FIG. 8 is a block diagram of a preferred embodiment of a finite impulse response filter of the present invention;

FIG. 9 is a graphical representation of the triangular response of the FIR filter of the present invention illustrated in FIG. 8;

FIG. 10 is a schematic diagram of the preferred embodiment of the partial coefficient generator section of the FIR filter of the present invention;

FIG. 11 is a schematic diagram of a typical selector logic block from FIG. 10;

FIG. 12 is a schematic diagram of the preferred embodiment of the initial section of the accumulator of the FIR filter of the present invention;

FIGS. 13A and 13B are a block diagram of the preferred embodiment of the remaining section of the accumulator of the present invention;

FIG. 14 is a schematic diagram of the timing signal generator of the present invention;

FIG. 15 is a schematic diagram of a typical DYN cell of the timing generator of the present invention;

FIG. 16 is a schematic diagram of the delay shift register cell of the accumulator of the present invention;

FIG. 17 is a schematic diagram of a typical inverter logic cell of the accumulator of the present invention;

FIG. 18 is a schematic diagram of the DAD cell of the present invention;

FIG. 19 is a schematic diagram of a typical SDL cell of the accumulator of the present invention; and

FIG. 20 is a schematic diagram of a typical DL cell of the accumulator of the present invention.

In general, as depicted in FIG. 1 in block diagram form, the present invention is an integrated analog-to-digital converter, generally designated 10. The converter comprises a feedback coder and a decimator, each of which is itself novel. The unique feedback coder of the present invention could be used to drive a conventional decimator. Alternatively, a conventional feedback coder could be used to provide the input to th eunique decimator of the present invention. However, using these novel components together provides optimum results and they are described together for that reason.

In its most preferred embodiment, the feedback coder takes the form of a single bit, second order delta sigma modulator 12 in which the filter elements are conventional CMOS push/pull amplifiers, as explained below. Modulator 12 receives the analog signal to be digitized at its input and generates a delta sigma stream (DSM) of unweighted bits representative of the amplitude of the received analog signal.

The output of modulator 12 forms the input to a decimator 14 which includes a partial coefficient generator 15 and a data accumulator 16. Generator 15 and accumulator 16 function as a finite impulse response filter and a subsampler. In its most preferred form, the finite impulse response filter has a triangular response. Generator 15 generates parallel partial coefficient words to accumulator 16. Accumulator 16 includes a simple multiplex circuit. The serial digital output of accumulator 16 provides the output of the converter. The output of the accumulator is a serial pulse code modulated word.

As will become apparent, the configuration of modulator 12, generator 15, and accumulator 16 of the present invention has been simplified, as explained in detail below, in order to eliminate the necessity for discrete components and to permit the entire converter to be fabricated on an integrated semiconductor chip of relatively small size.

The feedback coder of the present invention is represented in block diagram form in FIG. 2, in its most generalized form. The coder is illustrated as a generalized N-bit feedback coder, although the preferred embodiment described in detail below is a single bit feedback coder. The analog input signal X(t) forms the input to a summation circuit 20, the output of which is connected to a first filter G(S) 22. The output of filter 22 forms the input to an analog-to-digital circuit 24, the output of which forms the coder output Y(t), and the input to a feedback loop which includes a digital-to-analog circuit 26 and a second filter H(S) 28. The output of filter 28 forms the second input to summation circuit 20. A conventional circuit 30 is connected between the output of analog-to-digital circuit 24 and the control input of digital-to-analog circuit 26 to adapt digital-to-analog circuit 26 to the required step size.

Filters 22 and 28 shape the quantization noise spectra at the output Y(t) so as to move the dominant part of this spectra past twice the maximum input frequency X(t). The quantization noise can then be rejected by further filtering of Y(t) without affecting the encoded information signal X(t). If the transfer function of X(t) to Y(t) is not flat in the pass band, then inverse filtering of Y(t) is required to restore the signal.

Filters of this type are formed using passive or active (op-amp) components. The performance requirements for the active components are difficult to meet because of the high sampling rate of the system. In the prior art, these filters have been formed with only passive components or by switched capacitor circuits in MOS designs. However, forming the filters with only passive components restricts the possible functions which can be realized for these filters. On the other hand, using MOS switched capacitor circuits results in large process sensitive filters where the upper frequency limit of the sampling frequency is restricted. Further, the sampled data operation of the switched capacitor filters can degrade the preformance of the system.

The present invention overcomes the above disadvantages by substituting integrators including simple amplifier circuits as gain elements for these filters. FIG. 3 is a schematic diagram of a basic CMOS push/pull amplifier which, when used in conjunction with a feedback capacitor and the appropriate resistors, forms the filters in the coder of the present invention.

The circuit disclosed in FIG. 3 is a simple inverter, when used in the digital section of the system. It may be used as an active gain stage provided differential amplification is not required and the circuit can operate with large random and systematic DC offset voltages.

Differential amplification is not a problem in the feedback coder of the present invention which requires only single ended gain stages in the filters. The DC offset voltages have no deleterious effect provided that the DC transfer response to a DC error signal in the forward path is zero and the feedback filter 28 is realized with passive elements.

The circuit of FIG. 3 has a single dominant pole determined by loading at the output and the drain-to-source conductance (GDS) of the transistors. The gain is determined by the transconductance of the transistors and the drain-to-source conductance. Thus, the gain bandwidth can be made very high. The slew rate of the circuit is determined by the large signal current capacity of the transistors and can be made very large by using large device sizes.

Accordingly, the performance of the amplifying stage shown in FIG. 3 is well suited for the requirements of the feedback coder filter block. The amplifier allows improved performance of the generalized feedback coder when compared to specifications achievable by large MOS operational amplifiers.

The standard MOS process provides means of realizing passive resistors and capacitors. These elements are required in addition to the amplifiers shown in FIG. 3 for implementing the filter blocks for the feedback coder.

A problem with the passive components in MOS systems is the large component tolerance variation. This causes the variation in absolute pole-zole position of the feedback coder filters to be large. This variation, however, has only a small effect on the feedback coder performance.

A single bit feedback coder can be formed from the generalized N-bit coder of FIG. 2 by replacing analog-to-digital circuit 24 with a comparator 32 and a clocked latch 34, as shown in FIG. 4. In addition, the digital-to-analog circuit 26 can be replaced by a simple two position switch 36, clocked by the output of the coder.

Further simplification is possible by eliminating switch 36, such that the output of the latch is used to directly drive filter 28. This is achieved by designing the loop gain of the system to operate with reference voltages of V.sub.DD and V.sub.SS respectively.

A block diagram of a second-order single bit feedback coder which has a closed loop pass band transfer function of unity (also known as a second-order delta-sigma modulator) is illustrated in FIG. 5. It consists of first and second summation circuits 38, 40, a first filter 42, a second filter 44, a comparator 46, and a clocked latch 48. Each filter comprises a first order low pass filter, preferably in the form of a single order integrator, most preferably in the form of the CMOS push/pull amplifier circuit described above. The output of latch 48 forms an input for each of the summation circuits 38 and 40.

The details of the second-order delta-sigma modulator 12 of the present invention are illustrated in the schematic diagram of FIG. 6. In this circuit, further simplification is achieved by eliminating the comparator and performing the comparator function directly in a conventional static D-type flip-flop implemented in CMOS, as illustrated in FIG. 6A. The analog input signal is received at resistor 50 which, along with capacitor 52, form a single pole filter to remove all signal components above half the sampling rate of the delta sigma modulator. Resistors 54, 56, capacitor 58, and active gain element 60 (which takes the form of the amplifier illustrated in FIG. 3) form the first summer 38 and filter 42. The second summer 40 and the second filter 44 are formed in a similar manner using resistor 62, resistor 64, capacitor 66, and amplifier 68 (of the type shown in FIG. 3).

D-type flip-flop circuit 70 illustrated in FIG. 6A and inverter 72 replace comparator 46 of FIG. 5. The data input D of flip-flop 70 is connected to the output of amplifier 68. The clock inputs CK and CK of the flip-flop are fed with a timing pulse C.sub.p (which may be, for example, 2 MHz) from a baud rate generator (not shown) and the complement thereof. The Q output of flip-flop 70 forms the input to an inverter 72. (If a Q output is available, inverter 72 is not required.) The output of inverter 72 (or the Q output of the flip-flop 70, if available) is applied to the second summer 40 through resistor 64 and to the input of an inverter 74. The output of inverter 74 forms the input to the first summer 38 through resistor 56 and the input to an inverter 76. The output of inverter 76 is the output of the delta-sigma modulator, designated DSM. Inverter 76 may also be eliminated if the output of the modulator is taken from the output of inverter 72 (or directly from the Q output of flip-flop 70, if available).

The DSM output of the delta-sigma modulator is converted into a pulse code modulated (PCM) word by filtering it with a finite impulse response (FIR) filter and sub-sampling the output of the FIR filter at the PCM word rate. The combination of these functions is often called a decimator.

The number of filter sections and the value of the taps determine the response of the FIR filter. The filtering of the signal, in addition to forming the PCM data word, provides the anti-alias filtering to allow the output of the filter to be sub-sampled to provide a PCM data stream at a much lower rate than the delta sigma modulator sampling rate. The combination of FIR filtering and sub-sampling provide significant reduction in hardware requirements over the realization of the two block separation. A conceptual full hardware realization is illustrated in block form in FIG. 7.

Hardware reduction is possible when it is observed that the data output from the FIR filter is only required at the sub-sampling sampling interval. An example of how this hardware reduction can be achieved is set forth in block diagram form in FIG. 8. As seen in FIG. 8, the data input is applied to the input of the first of three tandem connected 64-bit shift registers 80, 82, 84 and to the address inputs of a read only memory 86. The outputs of each of the registers are also applied to the address inputs of the ROM, along with the output of a 6-bit binary counter 88. The data output of the ROM is connected to the input of an accumulator 90 which generates the data output.

In this implementation, sub-sampling takes place at a rate equal to one-sixty-fourth of the input rate and the FIR filter is 256 taps long. However, the structure allows any number of FIR filter tap positions. Tap size is altered by changing the size of the binary counter and the ROM size. The sub-sampling rate can also be adjusted by changing the number of shift register blocks used to delay the DSM data and changing the ROM size.

The simpler case of sub-sampling at the tap rates can be considered for a 64 tap decimator. A 6-bit binary counter 88 addresses ROM 86. The data address is the 7th bit of the ROM. If no data is present at the input, no data is added to the accumulator. If data is present, then the filter binary coefficient a.sub.n is down loaded from the ROM if the binary counter input to the ROM is c.sub.n. The data in ROM 86 is representative of the data for one of (in this case) 64 tap positions in the FIR filter. Hence, after 64 clock cycles, 64 data input points will have cycled through all 64 filter coefficients. After 64 clock cycles, the accumulator is dumped and cleared.

The data dumped from the accumulator is equal to the data at the output of the FIR filter in FIG. 7 at the sampling instant (the sampling occurs when the switch is closed). Note that the data at the output of the FIR filter in FIG. 7 is not used when the sampling switch is open. In effect, we have a serial implementation of the FIR filter and sampler shown in FIG. 7. Hence, the implementation illustrated in FIG. 8 is equivalent to the FIG. 7 implementation with significant reduction in hardware complexity.

Now consider the case of a sample every 64 data input cycles and a FIR filter of 256 taps long. In this implementation, the 256 coefficients of the FIR filter must be accumulated in 64 clock periods. This can be achieved by expanding ROM 86 to contain the coefficient data groups containing 16 possible outputs for every clock cycle. The 16 possible outputs represent the 16 combinations of four coefficients from the FIR filter, summed for the 16 possible data point combinations that appear at the ROM at a given binary counter period n. The contents of the ROM for the binary counter input C.sub.n are shown in the following table:

Using the approach outlined, all 256 positions will be multiplied by 256 data points in 64 time periods. In this implementation, each piece of DSM output data is sampled four times for four separate sampling periods. This method can be summarized as a serial-parallel implementation of the sampled FIR filter of FIG. 7. At each binary counter input to ROM 86, c.sub.n, four parallel data points are examined and over the 64 clock cycles, all 256 taps positions are utilized.

For the special case of a decimator composed of a FIR filter with a triangular impulse response and a sub-sampling rate of half of the FIR filter length, the ROM 86 may be replaced by simple combinational logic. FIG. 9 is a graphical representation of the impulse response transfer function for a 128 tap filter. The sub-sampling rate is one-sixty-fourth of the input rate for the FIR filter of length 128.

The simplification achieved by eliminating ROM 86 can best be understood by examining the nature of the ROM code for the system described and noting that for the triangular impulse response a.sub.n =c.sub.n+1 and a.sub.n+64 =63-c.sub.n. It is also noted that 63-c.sub.n =c.sub.n, where c.sub.n is the complement of each bit in c.sub.n. It is therefore possible to implement the function shown in the following table with the combinational logic illustrated in FIGS. 10 and 11:

Referring to FIG. 10, the DSM output from the delta-sigma modulator 12 is inverted by an inverter 100 and fed to the data input of a 64-bit shift register 102. Shift register 102 receives timing signals .0..sub.1, .0..sub.1, .0..sub.2, and .0..sub.2 from a timing generator (not shown) and generates at its output the delayed DSM signal, designated DDSM. DSM and DDSM are each separately inverted by inverters 104 and 106. The outputs of inverters 104 and 106 form the inputs to data selector logic circuits 108, one of which is schematically illustrated in FIG. 11.

Each logic circuit 108 receives a different timing signal from a timing generator described below and illustrated in FIG. 14. The timing generator is the equivalent of counter 88 of FIG. 8 and the timing signals are the equivalent of the output thereof, c.sub.n. These timing signals can be, for example, pulse trains having frequencies of 1 MHz, 500 KHz, 250 KHz, 125 KHz, 62.5 KHz, and 31.23 KHz, respectively.

Each logic circuit 108 receives one bit each of the DSM and DDSM signals at a time. The output of each logic circuit 108 (except the circuit receiving the 1 MHz timing signal) is inverted by an inverter 110 to form outputs S.sub.1. . . S.sub.6. The logic function of each logic circuit 108 and inverter 110 is as follows:

As will be appreciated from the above table, the value of the timing pulse forms the output if DSM alone is high. The complement of the value of the timing pulse forms the output if DDSM is high. A "1" forms the output if both DSM and DDSM are high. A zero forms the output if both DSM and DDSM are low.

The parallel combination of the six logic circuits 108 with the timing pulse trains generate a logic result equivalent to that set forth in Table II, with the exception that the second line gives a logical output of c.sub.n instead of c.sub.n+1. It is for this reason that the DSM signal itself forms a seventh output of the circuit of FIG. 10 to provide the additional required bit. The DSM output signal is re-named CIN and is the carry bit.

In conventional decimator structures, all of the bits of the parallel data word at the output of the ROM are applied to the accumulator at the delta-sigma input stream word rate. Such an accumulator design requires a very large chip area.

In the present invention, the accumulator is significantly simplified by using a pipeland approach to the accumulation process. The approach results in the unit cells running at the delta-sigma word rate with no requirement for the cells to propagate data through to other cells during each cycle. The pipeland approach also results in bits at the output of the accumulator arriving sequentially. Thus, a simple multiplexer can be employed to output the serial data words.

In order to pipeline the accumulator, the parallel input data word S.sub.1. . . S.sub.6 from the combinational logic of FIG. 10 is delayed bit by bit in a delay circuit, as shown in FIG. 12. As is clear from FIG. 12, the parallel outputs S.sub.1. . . S.sub.6 from FIG. 10 are each delayed by a number of bits which is a function of the position of the bit in the word. Thus, bit S.sub.1 is delayed no bits, bit S.sub.2 is delayed one bit, bit S.sub.3 is delayed two bits, bit S.sub.4 is delayed three bits, bit S.sub.5 is delayed four bits, and bit S.sub.6 is delayed five bits. The delay is achieved through the use of shift registers 112 of the required length. The carry bit output CIN is passed through the delay stage without delay. In the ROM implementation, there are no carry bits so that the CIN input of the accumulator would normally be left low. Outputs OS.sub.1. . . OS.sub.6 and CIN are fed to the multi-stage circuit shown in FIGS. 13A and 13B.

As shown in FIGS. 13A and 13B, the circuit has 12 stages, each stage including an adder, such as a full adder 114 (although half adders 116 may be used in the latter stages, if desired), a carry output delay logic cell 118, and a sum output delay logic cell 120, and an inverter logic cell 122, except for the last stage, the structure of which is discussed below.

The clock input to full adder 114 of the first stage receives the CIN signal. The clock inputs of the adder circuits in subsequent stages receive the output of the carry output delay logic circuit 118 of the previous stage. One data input of each of the full adders 114 in the first six stages receive, respectively, signals OS.sub.1. . . OS.sub.6 from the delay circuit of FIG. 12. In each stage, the second data input of the adder receives the output of the sum output delay logic circuit 120. The carry output of the adder in each stage is connected to the input of the carry output delay logic circuit 118. The sum output of the adder 114 is connected to the input of the sum delay logic circuit 120 and the input of the inverter logic cell 122, the structure of which is illustrated in FIG. 17. In the first stage, the data input to inverter logic cell 122 is a timing signal CLRIN from the timing signal generator (FIG. 14) which has been inverted by an inverter 124.

A 12-bit multiplexer comprising shift register bits 126 is aligned with the accumulator such that one shift register bit 126 corresponds to each stage of the accumulator. In the second and subsequent stages of the accumulator, the input of the shift register 126 is connected to receive the output of the sum delay logic circuit 120 from the previous stage. The outputs of the various inverter logic cells 122 are connected to an output line 128 which , in turn, is connected to the input of a delay shift register 130 (illustrated schematically in FIG. 16). The output of shift register 130 is inverted by inverter 132.

The sum output is fed back to the input of the adder of each accumulator stage and the carry output is fed to the adder of the next stage. Thus, if a data word was presented at the inputs of the accumulator, then the sums produced would not align with the carries generated. If the input data word is delayed bit by bit, that is, the first bit (LSB) incurs no delay, the second bit, one delay, et cetera, as in the delay circuit of FIG. 12, then the carry output from the first accumulator stage is delayed one bit and realigns with the input data bit which has been delayed at the input of the accumulator. In this way, it is possible to pipeline the accumulation process.

Upon the last input data word, the sum of the least significant bit is produced after one bit delay and the sum of the second stage produced after two bits delay, and so on. A simple multiplexer is used to select the valid output sum onto a single data output line 128.

The only timing required is given by a single clock pulse CLRIN which clears the sum of the first stage at the end of the accumulation cycle. This pulse is simply delayed one bit to generate a clear pulse for the second stage of the accumulator, etc. It should be noted that the full adders are only necessary for the six least significant bits of the accumulator and half adders for the most significant bits (alternately, a counter could be used).

The output from the accumulator is an offset binary data word. To convert the data word into two's complement format, the sign bit (i.e., the most significant bit) has to be inverted. This is accomplished by the delayed pulse CLRIN presetting the most significant stage of the accumulator. Thus, the output of the last shift register stage 126 is connected to the preset input of a DAD cell 134 (FIG. 18) in the last stage of the accumulator. It should be noted that the last stage of the accumulator has no carry output delay logic circuit 118 or sum delay logic circuit 120. Note that shift register 130 is a delay shift register and the output of inverter 132 is the serial output of the accumulator.

FIG. 14 illustrates the structure of a timing signal generator 150 which generates the various timing pulse trains utilized in the filter shown in FIG. 10 and the timing signal CLRIN used for clearing the shift register bits 126 of the multiplexer which forms a portion of the accumulator shown in FIGS. 13A and 13B. Timing generator 150 receives a master timing signal Cp (2MHz) which is the output of a baud rate generator (not shsown). The Cp signal is inverted by an inverter 152 and applied to bus 154. The output of inverter 152 is inverted in an inverter 156 and applied to a second bus 158.

The timing generator comprises a seven stage divider, each stage including a DYN cell 160, the structure of which is illustrated in FIG. 5. The CK and CK inputs to each cell 160 are connected to buses 158 and 154, respectively. The data input D of each DYN cell 160 receives the output of an EX NORgate 162 through an inverter 164.

V.sub.DD also forms the input to a NANDgate 166, the other input to gate 166 is the Q output of the DYN cell 160, upon which the 1 MHz output signal appears. The output of gate 166 forms one input to the EX NORgate 162 of the next stage. The output of gate 166 also is applied to one input of an EX ORgate 168, the other input of EX ORgate 168 is the Q output of the DYN cell 160 of the second stage upon which appears the 500 KHz signal. Each succeeding stage serves to divide the output of the prior stage in half. The CLRIN signal appears at the output of the next to the last stage.

It will now be appreciated that the present invention relates to an integrated analog-to-digital converter which can be formed on a semiconductor chip in a relatively small area and has reduced process sensitivity. The invention includes an integrated feedback coder which utilizes simple CMOS push/pull amplifiers in the gain stages so as to reduce the process sensitivity of the coder. A simplified finite impulse response filter employs a partial coefficient generator including a read only memory, counter and shift registers. The simplification occurs by observing that the data output of the FIR filter is not required at the input stream rate. A further simplification of the filter is employed for the special case of a triangular impulse response, such that the resulting PCM output stream is available at a rate equal to the PCM input stream rate divided by half of the number of taps of the windowing filter. The accumulator and shift register logic blocks of the filter are combined into a single block. In addition, by using a pipelined approach for the accumulator process, all stages of the accumulator run at the input word rate with no requirement for the stages to propagate data through more than one other stage during each cycle.

While only a limited number of preferred embodiments of the present invention have been disclosed herein for purposes of illustration, it is obvious that many variations and modifications could be made thereto. It is intended to cover all of these variations and modifications which fall within the scope of the present invention, as defined by the following claims: