Coherent Data Communications Technique

1. Field of the Invention

This invention relates to digital data communication, and in particular to a technique for coherently demodulating asynchronous phase modulated carrier signals transmitted through a noisy medium.

2. Prior Art

"Base band" (high and low or "non-return-to-zero") digital data signals may be communicated through a medium such as an AC power line by providing a carrier wave in the range of 100 kHz to 400 kHz, modulating the carrier with the data signals, and adding the modulated carrier to the transmission medium. Carriers have been modulated in various ways. Amplitude and frequency modulation (AM and FM) are common because they are relatively simple, but they have the drawbacks of high error rates and wide bandwidths. These drawbacks are minimized by phase modulation (PM), also known as phase shift keying (PSK).

Referring to FIG. 1, a carrier signal A may be phase modulated by combining the carrier signal and a base band data signal B through an exclusive OR (XOR) gate. At each transition in the data signals, the gate phase-shifts the carrier signal 180.degree. to produce a PSK modulated signal C. A phase shift indicates a transition between data bit values, but does not indicate the bit values themselves. The bit values are indicated by the length of time between transitions. For example, a high level data bit can be indicated by a short (high or low) signal, and a low level data bit by a long (high or low) signal. A receiver measures the time between carrier phase shifts to reconstruct the original data. Phase modulation is a conventional technique, but up until the present invention, it has required such elaborate circuitry for modulation and demodulation that it has not been economical. There remains a need for a better technique to communicate digital data signals in the presence of heavy noise, particularly random, spurious, and impulse type noises.

The present invention is an economical technique for narrow band-communication of digital data by phase modulation. In the invention, a receiver employs parallel channels driven by a multiphase clock to sample data, avoiding the need for a complex analog phase locked loop. A novel two-stage statistical filtering technique compresses sample values, finds their moving average, and applies hysteresis to resolve correct information, even using a very noisy carrier signal. A processor uses a pattern recognition algorithm to reconstruct the transmitted data. The invention is preferably implemented in integrated circuit form and thus is highly reliable over time and relatively immune to temperature effects.

FIG. 1 is a timing diagram showing a clock signal (A) which is combined with a data signal (B) to produce a PSK signal (C), and the bandlimited PSK signal (D) outlined by envelope (E);

FIG. 2 is a block diagram of the transmitter and receiver;

FIG. 3 is a schematic of the transmitter;

FIG. 4 is a schematic of the receiver;

FIG. 5 is a timing diagram showing the relationship between the received noisy carrier signal (F), the sampling clocks .phi.1 and .phi.2, and the bin clock .phi.bin;

FIG. 6 shows how at least one of the two sample clocks offset by 90.degree. will take samples in a maximum amplitude quadrant of the carrier signal;

FIGS. 7a and 7b show how three samples are counted to give one bin value;

FIG. 8 is a timing diagram showing the relationships of the received noisy data (G) signal to the samples (H), the samples to the bin values (I), and the bin values to the moving average window (J) of bin values (K);

FIG. 9A is a table of the transmit format expected by the pattern recognition, algorithm;

FIG. 9B shows the upper and lower statistical bounds of each received bit type.

FIG. 10 is a flow chart of the main routine by which the processor recognizes received characters; and

FIG. 11 is a flow chart of the subroutine by which the processor recognizes individual data bits and collects the bits to form characters.

Referring to FIG. 1, carrier signals A may be phase-shift modulated at predetermined time intervals to communicate information signals B. The main function of the present invention 10 is to recover a modulated carrier C and data signals B from a transmission medium with overwhelming spurious and random noise. The receiver of the present invention may be used for data which has been phase modulated and transmitted by any of several conventional multi-phase encoding techniques where the number n of phases is at least 2. Data bits are encoded as signals of several different lengths according to a format determined by the designer who

(1) selects transmit timing formats for "1", "0", and start/stop bits (in this example, high level data bits are represented by short duration signals, and low level data bits are represented by long duration signals);

(2) determines the statistical variance of the "1", "0" and start/stop bit probability density functions; and

(3) adjusts the transmit format so that the probability density functions of the transmitted/received data types are non-overlapping statistical domains.

FIG. 2 illustrates the invention with an improved and preferred transmitter circuit 12 suitable for two-phase transmission. Processor 14 generates a square wave carrier signal A and a digital data signal B, which are combined by XOR gate 16 to produce a PSK signal C as in FIG. 1. The PSK signal is then conditioned by an interface for the type of medium through which it is to be transmitted. Examples of transmission media include a coaxial cable data-link, an AC power line, and other well known media.

FIG. 3 details a transmitter preferred for use with carrier-current transmission medium. Signal C is bandlimited by an L-C bandpass filter 18, which smooths the square wave C into a sine wave D having reduced amplitude in the vicinity of phase shifts as outlined by envelope E. Typically, bandpass filter 18 will have a "Q" high enough to pass a relatively narrow bandwidth, reduce the carrier's high order harmonic components, and avoid causing sideband interference with radio frequency signals.

After filter 18, the smoothed PSK signal D is amplified by amplifier 20 to an appropriate level for transmission through a medium, such as an AC power line (not shown) connected at terminal 24 which may be an ordinary plug. A transformer 22 couples signal D to the power line for safety, but otherwise does not affect signal D.

PSK signals received from the transmission medium are first conditioned by an interface as necessary, then processed to recover the original data signals. The preferred embodiment of the receiver 40 in FIG. 4 has an interface 42 to tap signals from the transmission line, through, for example, an ordinary plug 44. The plug is connected in series to the ends of a coil 46 making up half of a coupling transformer 48. Transformer 48 is a high Q bandpass filter tuned to remove whatever frequencies predominate in the transmission medium, say 50 Hz, 60 Hz, and out-of-band noise, and to pass the carrier frequency. An inverted feedback amplifier 50 sets the transformer output signal 51 level to the threshold of amplifier 52. Transformer 48 produces, at its output line 51, a noisy analog waveform F as in FIG. 5. The analog signal F is squared up by a series of inverting amplifiers 54.

Waveform F is then sampled and filtered to reconstruct the transmitted data stream B. A multi-phase clock (MPC) generator 60 produces n sampling clock signals .phi.1, .phi.2, . . . .phi.n at the carrier clock signal A frequency fc .+-. an error of less than one quarter of a clock period during the number of clock cycles required for a character to be transmitted. In this example, a character requires up to 353 "bins", with 3 cycles per bin, giving a clock error tolerance of .+-.1/(353*4). The n clock phases are offset from each other by 360.degree./2n. The number of sampling clock phases used is equal to the number of phases, n, of the PSK signal C. There are two reasons for using a multi-phase sampling clock.

First, a mismatch between independent transmitter and receiver clocks destroys good data. Multi-phase-clocks and duplicate channels operating in parallel solve this problem. If one of n sampling clocks samples in a minimum amplitude quadrant of the carrier signal, at least one other sampling clock phase offset by 360.degree./2n samples in a maximum amplitude quadrant.

Second, with a MPC there is a higher probability of receiving valid information in the presence of random noise. In conventional systems some lower limit or threshold, for example zero dB s/n, must be established to evaluate the received information. Information below the threshold is typically lost. In contrast, the present invention uses multi-phase-clocks which increase the probability of detecting correct signals in a poor signal-to-noise ratio environment.

In the example of two phase modulation, the MPC 60 generates two sampling clocks .phi.1 and .phi.2 with a phase difference of 90.degree., and a bin clock .phi.bin at one-third the sample clock frequency. FIG. 6 shows how sampling a sine wave, like the smoothed PSK signal, twice per period with the sample clock periods offset 90.degree. insures that at least one of the samples will fall during the two quadrants of positive and negative maximum amplitude, 45.degree. to 115.degree. and 225.degree. to 315.degree.. The signal is sampled at 70.7% or more of its maximum cycle amplitude point for whatever actual signal levels are received. If for example the incoming signal has deteriorated to 10% of its maximum transmitted amplitude, one of the MPCs will sample at the 70.7% or greater level of the 10% signal.

Referring again to FIG. 4, MPC 60 uses a crystal, appropriate connections, and a counter 62, preferably a type HC161, to divide the crystal frequency down to a 1 MHz reference frequency. A latch 64, preferably a type C374, and a ROM 66, preferably a type S288, divide the reference frequency into the two sample clock signals .phi.1 and .phi.2 at the same frequency but offset from each other by 90.degree., and a bin clock .phi.bin, FIG. 5, at one-third their frequency.

The received signal F is processed by parallel channels I and II, which are identical except for being driven by the offset sample clocks. For clarity's sake, only one of the channels will be described. Even though at least one of the samples in each MPC cycle is sure to be in a maximum amplitude quadrant, the carrier signal is generally dominated by clicking, spurious, impulse, and random noise, which the MPC cannot eliminate. All of the samples are liable to include a noise component several times larger than the data component. A two stage statistical filter was developed to contend with the noise problem.

In the first filtering stage, bit quantization, the signal at amplifier output terminal 70 is applied to the D input of a flip-flop 72, preferably type 4013. The flip-flop samples the data signal G (FIG. 2) and produces a string of samples H. Three consecutive values H are applied to an up-down counter 74 to arrive at a "bin" value I. On every third sample clock signal, bin clock .phi.bin resets up/down counter 74 to zero. The up/down counters may be type 4516. The designer may modify the number of samples-per-bin to accommodate more degrees-of-freedom than specified by the Central Limit Theorem and chi-square statistics. Quantizing three samples of data H into one bin value I provides hardware simplicity, a compression of three samples of information into one, and a more accurate statistical representation of the noise infested signal G than a 50/50 decision on one or two samples. FIG. 7a shows the eight possible sequences of three samples-input to up/down counter 74.

A "one" sample counts as a positive one, and a "zero" sample counts as a negative one. The sign of the sum gives the bin value. Referring to FIG. 7b, a zero MSB indicates a positive sum, and a bin value of 1; a one MSB indicates a negative sum, and a zero bin value. The final quantized decision level is based upon a 2/3 or greater occurrence of either logic level. Alternate decision levels and quantizations could be used instead at the designer's discretion. A four sample quantization would produce 2/4 bounds, and a five sample quantization would provide 3/5 bounds.

As shown in the FIG. 9 table, the number of bins transmitted is indicated by the timing between phase shifts, and the timing between phase shifts indicates the original bit values. Phase shifts are transitions of bin values from 1 to 0, or 0 to 1, depending on where the clocks .phi.1 and .phi.2 happen to be sampling relative to the period of carrier signal A. Noise considerations aside, if the two channels sample on opposite sides of the carrier signal zero crossings, then the two channel's bin values will be opposite.

The second stage of statistical filtering simulates the Central Limit Theorem by finding the moving average of a number of bins "N" according to the formula ##EQU1## where "x" is the value "1" or "0" in bin number "n". The designer specifies a statistical outcome to determine the number of bins, eleven in the preferred embodiment, in the moving average window J (FIG. 8). Many of the transitions between bin values I are random and are caused by noise rather than carrier phase changes. The eleven bin moving average technique suppresses or damps these random transitions to produce the string of bin values K. The moving average is found by a shift register and a ROM or a PROM 78, preferably a type DM87S185. The .phi.bin signal also clocks the bin value at the Q output terminal of up/down counter 74 into an eleven bit shift register 76. The register shifts in each new bin value I, and shifts out the eleventh oldest bit. Shift register 76 may be constructed using two type C164 eight bit shift registers, both incremented by the same .phi.bin pulse, and the discards from one register being used as the input to the other register. The PROM code for providing the moving average is listed in Appendix A.

In many situations the statistical degrees-of-freedom required may make the moving average technique costly to implement and so impractical that it would never approach its theoretical accuracy. A practical solution is to incorporate into the second stage of filtering a digital hysteresis with thresholds set at a small percentage of the statistical variance of the samples due to noise. The digital hysteresis logic for the carrier current embodiment is implemented by having PROM 78 output a binary "01" when 7 or more bin values are low, and a binary "10" when 7 or more bin values are high. In between, when there is a difference of 3 or less in the count of high and low bin values, PROM 78 outputs a "00". These values are applied to the input terminals of a JK flip-flop 80, preferably type 4027. With unlike input signals, the flip-flop follows the value at the J input terminal. With zeroes at both terminals, the flip-flop retains its previous state. J-K flip-flop 80 delays switching from one value to the other value for the eleven bin binary average until more than 6 bins have the new value. Thus the received signal is recovered substantially free of random noise contamination or signal ambiguity.

The next portion of the receiver circuit performs edge detection. The Q output signal of JK flip-flop 80 is applied to both input terminals of XOR gate 82. One of the Q output signals is passed through a DQ flip-flop 81 before being applied to the XOR gate. When the output of JK flip-flop 80 changes, the signal applied directly to XOR gate 82 will arrive slightly before the signal applied indirectly, so the XOR inputs will momentarily be unlike, and the gate 82 output will be high until the second signal arrives. Phase transitions or edges are well preserved by this statistical filtering.

Once a phase transition has been detected, pattern recognition is applied to bin cluster counts to determine whether the received bit is a "0", a "1", or a start/stop. FIGS. 9a shows the selected number of information bins transmitted for each bit type used in the carrier current example, and FIG. 9b shows the upper and lower statistical bounds of each received bit type. This example uses seven data bits and an eighth odd parity bit.

Pattern recognition determines the time interval, or number of bins, between a change in bin values K, and thus between phase shifts in the data modulated carrier signal C. The signal from the edge detection XOR gate 82 is applied to the reset terminal of a counter 84, preferably a type 4520. Bin clock .phi.bin increments the counter. Reset signals are also applied to a latch 86, preferably a type C374, to latch the count.

The bin counts approximate the time intervals between carrier phase shifts in the transmission. The pattern recognition algorithms in the FIG. 9 and 10 flowcharts analyze the count of bins to reconstruct the transmitted data string. The bin count analysis may be done on any suitable microprocessor, such as an INS 8070 using the program in Appendix B. For an alternate embodiment of the PSK system with n>2 phases the algorithms should be expanded.

The microprocessor 90 first initializes counters for several variables used in analyzing the bin counts. "CLCT" is a "cluster count" of the number of bins as received from latch 86; "OLDCLCT" is the old cluster count; "WBCT" is the "word" bit count; "ONECT" is the number of "1"s counted in a word; and "SYNCTR" is the number of channels that have analyzed a complete word of bits. The values for these variables depend upon the choice of communication format (number of bits-per-word and use of start-stop bits) and transmission format (number of bins to represent a "1", a "0", and a "start-stop"). The FIG. 10 flowchart is for the transmit format shown in FIGS. 9a and 9b. The variables CLCT, OLDCLCT, and ONECT are each initialized to zero and WBCT is initialized to nine.

Microprocessor 90 sets a "ready" flag to false (0) for each channel, and waits for an interrupt signal from either channel. The same XOR gate 82 signal that latches latch 86 and resets counter 84 also interrupts microprocessor 90 to notify it that a count is ready to be analyzed.

The duplicate sampling channels I and II both interrupt microprocessor 90, which executes a routine for each channel. Suppose that channel I is the first to complete a bin count in latch 86 and to send an interrupt signal to microprocessor 90. The microprocessor begins executing the routine in the FIG. 10 flowchart to analyze the cluster counts from latch 86. It finds errors and reinitializes the counters until it has recognized a correct number of bins and bits for a word, and then sets the "ready" flag to true (1). When the ready flag is set to true, the microprocessor routine in the FIG. 9 flowchart for that channel exits the waiting-for-ready loop, increments the SYNCTR counter, and waits a predetermined length of time for the other channel to successfully analyze a word, set its ready flag to true, and increment the SYNCTR counter. The START and END CHARACTER TIME OUT sequence near the end of the FIG. 10 flowchart is to prevent waiting indefinitely to receive data from a dead channel. The possibility of a dead or information-less channel is a function of the beating associated with transmitting and receiving data with asynchronous clocks. When the SYNCTR count equals 2, or when the length of time during which the other channel could recognize the same character has passed, microprocessor 90 goes on to "write" the recognized character to another device, not shown. The preferred embodiment of the invention, operating with a noisy 0 dB s/n power transmission line, successfully recognizes a character on one channel or the other about 4/5 of the time; on both channels about 1/5 of the time, and infrequently makes the mistake of "recognizing" two different characters. After a character is written, the processor resets the counters and ready flags as explained, and waits for another interrupt. Meanwhile, counter 84 will have been accumulating a bin count for the next data bit transmitted.

Details have been disclosed to illustrate the invention in a preferred embodiment of which adaptations and modifications within the spirit and scope of the invention will occur to those skilled in the art. The scope of the invention is limited only by the following claims. ##SPC1##