WO1993011618A1

WO1993011618A1 - Feedback communications link controller

Info

Publication number: WO1993011618A1
Application number: PCT/AU1992/000633
Authority: WO
Inventors: Stephen Clive Cook; Jason Beaufort Scholz
Original assignee: The Commonwealth Of Australia
Priority date: 1991-11-27
Filing date: 1992-11-26
Publication date: 1993-06-10

Abstract

The transmitter (1) transmits signals in the form of symbols to a receiver (2). The received signals are analyzed and a probability density function estimate is formed in a pdf estimating means (4). There is a memory means (5) which contains a library of known probability density functions. A comparison means (6) compares the measured histogram from the pdf estimating means (4) with those stored in the memory means (5). The comparison means (6) produces an index (8) which uniquely identifies the library pdf most closely matching the measured pdf. The index (8) is fed back. At the transmitting end of the communication link the index (8) is used to select the corresponding pdf from memory means (7). Memory means (7) and memory means (5) are equivalent. The pdf extracted from memory means (7) is used to adjust the transmission parameters of the system.

Description

FEEDBACK COMMUNICATIONS LINK CONTROLLER

BACKGROUND OF THE INVENTION

This invention relates to a method and apparatus for improving communication system performance where a feedback path is available, based on estimation of the statistical profile of the decision variables. The invention is in particular applied to digital communication systems, especially those required to operate over time-varying channels (ie HF skywave, VHF meteor burst, VHF line-of-sight, VHF troposcatter, satellite, mobile radio, military channels).

A transmitter as part of a digital communication system, sends information by representing that information as a sequence of symbols from a finite set or alphabet. This information may be corrupted by noise or other disturbances whilst traversing the communication channel. The decision variable is the signal or value derived in all digital receiving systems just prior to deciding which symbol was sent. The decision variable is. the only entity used in the receiver to decide what information was originally transmitted. The corruption of the information in the channel manifests itself as a perturbation in the decision variable from its expected value. Over a sequence of symbols, this may be visualised as a spread in decision variable values. The decision variable probability density function (pdf) is a statistical representation of the exact nature of this spread.

In a previous patent application (PCT/AU90/00581 ) a method of determining an error rate from a measured estimate of the decision variable pdf was described. What is now described is the means by which the pdf can be used to optimise digital communication link performance. Such an arrangement has particular advantages for controlling the errors in a digital communication link whilst allowing some other criteria to be optimised. For example, information throughput may be maximised whilst providing a guaranteed error rate performance. The invention is also useful for extending the usefulness of the communication link through difficult channel conditions (such as severe fading, noise or interference).

The invention can be said to be a method for adaptive link control (ALC). The ALC technique centres on the use of the estimated link pdf. The pdf gives far greater insight into current link performance than a simple error rate measure. This ϊs because the pdf is a more information-rich description of the communication link. Knowledge of the profile or shape of the pdf may be used to predict link performance if transmission parameters are altered. For example, if channel conditions were slowly varying, and transmit power were variable at the transmitter, the current pdf profile would allow calculation of how an increase (or decrease) in power would effect the occurence of received errors, thus allowing, say transmit power to be minimised whilst maintaining a ceiling on error rate performance.

In effect, almost all communication links utilize time-varying channels. To optimise link performance on a time-varying channel, the channel state must be measured, tracked over some period and then predicted. The time required to measure, estimate and predict the nature of the pdf, devise the best adaptation strategy and engineer the change must be somewhat faster than the rate of change of the channel conditions. Many channels change state quickly requiring a rapid means of measurement. In copending PCT/AU90/00581 a technique for pdf estimation requiring a very small number of samples is disclosed.

The technique of pdf estimation requiring a very small number of samples has performance equivalent to the pdf estimation technique of PCT/AU90/00581 with an infinitely large number of bins (or regions) for histogram formation.

instead of placing samples of the decision variables in bins or regions and forming a histogram, the new technique uses all samples in their continuous form. The basis of the technique is derived as follows:

The objective is to estimate which decision variable pdf model M_k (k=1 ,2,...,L) best represents the measured samples of the decision variable. According to Bayes rule, the a posteriori probability of link model Mk given the decision variable samples Yf (i=1 ,2,...,N) is :

P fM I Y Y \ _ P(Yι . Y2. ~. Y_N | M_k) P(M_k)

P (MR I Yi . Y₂. ... YN) - p(Y₁ , Y₂. ... Y_N) ^{(1 }}

If the decision variable samples are independent and identically distributed, then:

p(Yι, Y* ... YN I Mk) = P(YιlMk) P(Y₂|M_k)_.... p(Y_N|Mk) (2) Substituting equation (2) into (1 ) gives:

N p(M_k)ri ( ii^M )

P(M_k l Y₁. Y₂, ... Y_N) = _p(Yι ^_{| ,}.. y_{N )} (3)

In order to estimate the best representing model, we wish to maximise the a posteriori probability (3) over all possible models k=1 ,2 L. Since p(Yι ,Y2.— Y ) is independent of k, it may be neglected. A further simplification results by taking the logarithm of (3) since in maximising a function, it is equivalent to consider maximising the logarithm of that function. This gives a Maximum A-posteriori Probability (MAP) estimator:

N p(k) = ∑ ln p(Yi| M_k) ₊ ln P(M_k) (4) i=1

If all models are equally likely, this reduces to a Maximum Likelihood (ML) estimator:

N p(k) = ∑ ln p(Yi|M_k) (5) i=1

Thus, we choose a model k = such that p( k ) is the maximum value over all k.

In order to evaluate the ML estimate of (5), the probability density function for model k must be evaluated at each sample value point Yj. Where an analytical expression for p(Yi|M_k) is known, (5) may be evaluated by substituting Yj values into the equation for p(Yj|M ) and calculating ρ(k). If there is no known analytical expression for p(Yj|M ); values must be stored as individual points for a finite set of values and the pdf value at the point Yj calculated by interpolation between the known points on the pdf. For smooth pdf functions only a small number of points needs to be stored to derive an accurate value at the measured point.

A communication link is normally considered to consist of an encoder, a modulator, a transmitter, a channel, a receiver, a demodulator and a decoder. The measurement of the decision variable usually occurs in the demodulator. The decision variable is usually the output of matched filters. For example, the phase of a phase-shift keyed (PSK) received signal. For quadrature amplitude modulation there are two dimensional decision variables associated with phase and amplitude.

It is known that the estimated pdf can be used to determine various channel parameters such as error rate. This only utilises a small aspect of the total amount of information carried in the pdf.

in the past, to optimise link performance using a feedback link, various parameters have been fed-back such as the signal-to-noise ratio or signal amplitude. These parameters provide only a narrow glimpse of the link condition whereas the pdf provides a substantially more detailed picture.

The transmission parameters which can be adapted include:

Transmit power

Data Rate

• bandwidth

• constellation packing Interleaving length

Selective channel diversity

• frequency

• space (antennas) Class of modulation scheme Detector Structure

Error Control Code

• Automatic Repeat Request Strategy

• Forward Error Correction Strategy

- code rate Protocols

• Preamble

• Packet length and structure Synchronisation Scheme Security level

It is desirable to send a minimal amount of information on the return channel.

Clearly the problem with feeding back the entire pdf is that a high bandwidth return channel is required to transmit sufficient data to uniquely map the receiver-measured pdf. This can introduce a new source of error as well as incurring a large communications overhead.

The inventors have found that it is not necessary to feed-back the entire pdf and indeed the transmitter need only be instructed by the receiver as to the appropriate next transmission parameter state. Either of two methods may be applied:

Firstly, it is sufficient to feed-back a pdf index which will be used to select a pdf from a range of forms stored in a memory means at the transmitter. Essentially, a library of probability density functions is already maintained at the receiver for the classification of the measured decision variable values. An identically indexed library is also maintained at the transmitter. The pdf is estimated using the histogram-based technique of patent PCT/AU90/00581 or preferrably using the technique described above in this patent. For channels with a high rate of change of conditions and / or large delays, the future pdf may need to be predicted; regardless, an index to the expected pdf is then fed-back from the receiver to the transmitter and the corresponding pdf is extracted from the library at the transmitter. The pdf can then be used to calculate modifications to the transmission parameters to suite the link condition represented by that pdf and the error requirements demanded by the user. Once the pdf is indentified, and a change in transmission parameters is to take effect, both transmitter and receiver must synchronise to the new configuration.

A second method can be used if the number of transmission parameter states is smaller than the number of pdf indicies. This method is to send an index to the transmission parameter state. For example, an adaptive feedback system uses a choice of two classes of codes, each of which has a choice of six code rates; thus there are twelve index values or transmission parameter states required (4 bits). The advantage of this method if that the transmitter does not need to run the algorithm to translate the pdf index into the best choice of transmission parameter. This is performed only once at the receiver end of the link. In most practical cases, the number of transmission parameter states will be much less than the number of pdf indicies, thus allowing a smaller amount of feedback information.

In certain situations, in particular when the link being controlled forms part of a network, there are definite advantages in feeding back an index to the decision variable pdf rather than the transmission parameter state. As has already been stated, the pdf provides a comprehensive description of the state of the link over the period of its estimation and as such contains far more information than transmission state. There are situations where the transmitter is required to have detailed knowledge of link performance both currently and historically and this can be derived readily at the transmitter from the sequence of pdf's indexed by the receiver. For example a network controller might wish to interrogate a transmitter to establish the probability of correct transmission of a given message, whether a link is degrading, or a complete set of transmission statistics such as average probability of error, average error burst duration, mean error free time, etc.

SUMMARY OF THE INVENTION

Therefore in one form, the invention can be said to reside in a method of optimising the performance of a communication system comprising the steps of : receiving signals at a receiving means; forming decision variables from each received signal; comparing over time the decision variables with values derived from a plurality of models maintained in a library; calculating a likelihood value for each library model said likelihood value being a measure of the degree to which the model agrees with the measured decision variables; selecting the model with the maximum likelihood value; utilizing the selected model to establish criteria for modifying a transmission parameter state of a transmitting means; sending back to the transmitting means an index uniquely specifying the next transmission parameter state; receiving the index at the transmitting means and using the index to select the next parameter state for the transmitting means.

In preference the model with the maximum likelihood value is selected by determining the logarithm of each of the likelihood values, comparing the calculated logarithm of the likelihood value for each library model and choosing the model associated with the largest log-likelihood value.

In preference there is a library of possible transmission parameter states and preferably the library of possible transmission parameter states is contained in a memory means and an identical library is provided at the transmitting means and at the receiving means. In this way a low bandwidth channel can be used to feedback the index to the transmission parameter state rather than the large bandwidth channel which would be required if the entire pdf were fed-back and the transmission parameter state calculated at the transmitter.

In one form the index identifies a memory location in the memory means. The memory means may be a ROM or EPROM or similar memory device. The index may then be an address offset.

In preference an analysis means determines the appropriate transmission parameter state to be employed by analysing the communication system requirements and the probability density function of the decision variables.

It will be realised that it is not essential to have a transmission parameter state library at the transmitting means. Since each index uniquely defines a pdf, all the necessary channel information is embodied in the index. Analysis means may therefore be devised which directly use the index as input.

The criteria for modifying the transmission parameters can be any of a wide range of established algorithms that calculate the appropriate changes to, for example, improve the error performance, increase the date rate, change the symbol set etc.

In another form of the invention there is proposed a method of assessing and optimising a communication system including the steps of : measuring at least one probability density function of transmission through an associated communication link or channel; comparing the measured probability density function with a library of probability density functions and selecting an index associated with the library probability density function most nearly matching the measured probability density function; transmitting the index from the receiver to the transmitter; and modifying transmission parameters of the transmitter according to criteria established with the index.

In preference the index identifies the location of a probability density function in a first memory means associated with the receiver and in a second memory means associated with the transmitter.

Alternatively, there is provided a calculating means at the receiver which is adapted to calculate the transmission parameters required to optimise the communication. In this case all that is required to be fed back to the transmitter is the next transmitter parameter state. This can either be achieved by feeding back the complete parameter state or an index to a library of parameter states. In the latter case there are memory means at the transmitter and receiver which contain a library of transmission state parameters.

In yet another form, the invention can be said to reside in a method of optimising a link of a communication system including the steps of : receiving a first signal; categorising at repetitive points in time a decision variable into a number of states; comparing over time the decision values with values derived from a plurality of models maintained in a library; calculating a likelihood value for each library model and choosing the model having the maximum likelihood value; utilizing the selected model to establish the criteria for modifying transmission parameters of the transmitting means; sending back to the transmitting means an index uniquely specifying the next transmission parameter state; receiving the index at the transmitting means; and using the index to select the next parameter state for the transmitting means to optimise the communication system.

In a further form, the invention can be said to reside in a method of optimising a link of a communication system including the steps of : receiving a first signal; categorising at repetitive points in time a decision variable into a number of states; measuring a probability density function for the transmission of a symbol or symbols over a communication channel or link; comparing the measured probability density function with a first library of probability density functions; selecting an index associated with the library probability density function most nearly matching the measured probability density function; transmitting the index from the receiver to the transmitter; using the index to select a probability density function from a second library,^' said second library being identical to said first library; and modifying transmission parameters of the transmitter according to criteria established with the probability density function to optimise the communication system.

In preference, the step of measuring a probability density function is characterised by recording as a histogram value the occurrences or proportion of occurrences over a number of symbol transmissions that the decision variable is categorised as being a particular category; and the step of comparing is further characterised. by the library probability density functions being for the same number of categories as the measured probability density function.

It will be appreciated that the library probability density function may be stored as a greater number of categories. However, when the step of comparing is conducted two or more stored categories are combined so that the step of comparing is between probability density functions having the same number of categories.

In a still further form the invention can be said to reside in an apparatus for optimising digital communication systems including: a bi-directional communication link adapted to transmit and receive a plurality of signals in the form of symbols; measuring means adapted to measure a probability density function associated with the link; memory means containing a library of known probability density functions; comparing means adapted to compare the measured probability density function with the library of known probability density functions and produce an index to the library probability density function which most closely matches the measured probability density function; feed-back means adapted to feed back the index; analysis means adapted to extract from a second memory means the probability density function indicated by the index and to determine appropriate transmission parameters to optimise the digital communication system.

Alternatively the invention can be said to reside in an apparatus for optimising digital communication systems including: a bi-directional communication link adapted to transmit and receive a plurality of signals in the form of symbols; measuring means adapted to form decision variables from each received signal; memory means containing a library of known probability density functions; comparing means adapted to compare the measured decision variable with values derived from the library of known probability density functions and produce a likelihood value, said likelihood value being a measure of the degree to which the known probability density function agrees with the measured decision variables and to select the probability density function associated with the maximum likelihood value; calculating means adapted to utilize the selected probability density function to calculate transmission state parameters required to optimise the communication system and to produce an index to a library of transmission state parameters; feed-back means adapted to feed back the index to the transmitter; memory means at the transmitter containing a library of transmission state parameters; and transmitter control means adapted to receive the index, select the transmission state parameters from the memory means according to the index and apply the parameters to the transmitter to optimise the communication system.

In yet another form the invention can be said to reside in apparatus for use in a digital communication system including: first transmitting means adapted to transmit via channel means signals in the form of symbols; first receiving means adapted to receive said signals; demodulator means adapted to form at least one decision variable from a received signal, each decision variable being characterised by a decision attribute which is useful in determining which of a plurality of possible symbols is being received; first memory means containing a library of probability density functions; calculating means adapted to produce an index; second transmission means adapted to transmit an index signal corresponding to said index; second receiver means associated with the first transmitter means and adapted to receive the transmitted index signal; second memory means containing a library of transmission state parameters; and transmitter control means adapted to extract transmission state parameters from the second memory means, according to the index signal, and to control the transmitter to transmit according to the selected transmitter state parameters to optimise the digital communication system.

In preference, the calculating means is adapted to: form at least one measured probability density function of the value of the decision attribute of an associated decision variable in histogram form by repetitively categorising the associated decision variable into categories dependent upon the value of the decision attribute, where the categories consecutively divide the range of values that the decision attribute can take, and each category has an associated range being a sub-range of the range of values that the decision attribute can take; record for each category the number of times or proportion of times that the value of the decision attribute is within the associated range of the category or the value of the decision variable is within or exceeds the region of the category; compare the measured probability density function with the probability density functions contained in the first memory means; and to provide an index associated with the library probability density function which most closely matches the measured probability density function.

Alternatively, the calculating means is adapted to compare over time the decision variable with values derived from each of the library probability density functions contained in the first memory means; calculate a likelihood value for each library probability density function, said likelihood value being a measure of the degree to which the value derived from the library probability density function agrees with the measured decision variables; and provide an index associated with the library probability density function having the maximum likelihood value.

In preference, the calculating means is further adapted to categorise the decision variables at the same time that a demodulator demodulates the decision variable into the symbol being received.

Time varying communications links may often be characterised by a relatively long time constant variation and a random (or pseudo-random) noise component. The time varying component can be determined and corrected for by using the pdf feedback described above. The noise component can also be estimated using similar techniques.

In another form of the invention there is proposed a method of optimising time- varying communication systems comprising the steps of repetitively measuring a decision variable, building a first histogram of decision variable measurements using a slow time constant and building a second histogram of decision variables using a fast time constant, comparing the histograms with a library of histograms and selecting respective indices of the histograms which most closely match the measured histograms, feeding back the selected indices, and adjusting transmission parameters of the system in accordance with criteria established with the indices.

For a better understanding of this invention a preferred embodiment will now be described with reference to the attached drawings in which :

FIG. 1 is a schematic of a bi-directional communications link;

FIG.2 is a schematic of a bi-directional communications link utilizing t e invention;

FIG.3 shows in block diagram form a digital receiver exhibiting aspects of the invention;

FIG.4 shows a sketch of how a decision variable may be categorised, showing an ideal receive waveform (decision variable) and what might be received;

FIG.5 illustrates a probability density function and a histogram estimation of it using the categories as illustrated in FIG. 4;

FIG.6 illustrates in block diagram form the selection of which stored probability histogram is best estimated by the measured histogram according to a first preferred embodiment;

FIG.7 shows the time varying form of a communication channel;

FIG.8 shows an embodiment of the invention which might be utilized in association with the channel of FIG. 7;

FIG.9 shows -a library of possible probability density functions;

FIG. 10 shows an embodiment useful in a communications link utilizing modems; and

FIG. 11 shows the performance of the embodiment of FIG. i 0. Referring to FIG. 1 there is shown a brief schematic of bi-directional communication system. The system consists of a transmitter 1 , a receiver 2 and a channel 3. In a bi-directional system there is facility for feedback to occur between the receiver and the transmitter. This allows for the possibility of the receiver providing information on how well it is receiving the transmitted information and/or to what degree the channel is corrupting the transmitted signal. To reduce communications overhead it is desirable to keep the bandwidth of the feed-back channel to a minimum.

FIG. 2 shows the communication link of FIG 1 with the addition of the pdf feedback invention. The transmitter 1 transmits signals in the form of symbols to a receiver 2. The received signals are analyzed and a probability density function estimate is formed in a pdf estimating means 4. There is a memory means 5 which . contains a library of known probability density functions. A comparison means 6 compares the measured histogram from the pdf estimating means 4 with those stored in the memory means 5. The comparison means 6 produces an index 8 which uniquely identifies the library pdf most closely matching the measured pdf. The index 8 is fed back. At the transmitting end of the communication link the index 8 is used to select the corresponding pdf from memory means 7. Memory means 7 and memory means 5 are equivalent. The pdf extracted from memory means 7 is used to adjust the transmission parameters of the system.

Referring to FIG. 3, here the digital communication receiver is illustrated with the invention applied to it. This receiver consists of an antenna 9, RF (radio frequency) stage 10, demodulator comprising a decision variable extractor 11 and decision maker 12, decoder 13, decision variable 14, and further a categorising and counter stage 15, and a computing and comparison stage 16. The computing and comparison stage 16 provides the index 8 which is fed back to the transmitter 1. The memory 17 provides a library for use by the computing and comparison stage 16.

The antenna 9 picks up transmissions which are then amplified and conditioned in the RF stage 10. The demodulator reduces the received signal ready for the decoder 13 to decode the symbol sent. The decision variable extractor 11 forms one or more decision variables from the received signal. The decision maker 12 provides an output for the decoder 13 to decode. The output signal 18 of the decoder 13 is often characterised, in a binary system, as being a series of discrete bits. To obtain a measured histogram to estimate the probability density function of the communication link, the decision variable is categorised into a number of categories. To explain this by example, FIG. 4 illustrates an ideal waveform indicated by circles (examples being shown by 19) of a decision variable as would be received if there was no disturbances in the communication link and the waveform indicated by crosses (example being shown by 20) that might be received in atypical operating system. Here the waveforms 19 and 20 are both forms that the decision variable may take.

The decision variable ideally has values of zero "0" or one "1". The value of the decision variable is categorised into categories a, b, c, d,e, f, g, h i, j, k, I, m, n, o illustrated by arrow 21.

Within a demodulator the decision variable is sampled in response to a clock signal at points marked by arrows 22. This is emulated whilst obtaining a histogram for the communication link probability density function with the value of the decision variable being categorised into which of the categories it falls within at the time of the clock signal.

Over a number of symbol periods, one being illustrated by arrow 23, the count within the categories 21, estimates the communication link probability density function. This is illustrated in FIG. 5 where the histograms 24 and 25 approximate the probability density functions 26 and 27 respectively for the example illustrated in FIG. 4. The histogram is an accumulation of occurrence of decision variables being categorised in a particular category.

In FIG. 6 which illustrates the selection of which stored probability histogram is best estimated by the measured histogram according to a first preferred embodiment, the decision variable 14 is applied to a number of threshold detectors 28, 29, 30. Each threshold detector is connected to a respective counter 31 , 32, 33. It will be appreciated that threshold detector 30 and counter 33 are the last of a series of such devices. The more threshold detectors then the more categories into which a decision variable is categorised and the greater accuracy that may be achieved. It has been found that a series of four categories which in turn means 4 threshold detectors and four counters will provide an estimate of the bit error rate down to 10" bit error rate for a 2400 bit per second link using 4000 samples with performance comparable with known bit error rate monitors using 100000 samples. The outputs of the counters 31 , 32, 33 are compared with a library of known probability density functions stored in memory means 17. These are compared by calculating means 34 which may be a computer. The output of the calculating means 34 is the index associated with the stored probability density function selected as most closely resembling the measured histogram according to the counts in the counters 31 , 32, 33.

The embodiment of FIG. 6 provides in the counters a category cumulative estimation of the probability density function of the link. This is because the use of simple value exceeded threshold detectors instead of threshold detectors which detect if the decision variable falls within a range of values. The calculating means 34 may convert the category cumulative estimation of the probability density function to a category non-cumulative probability density function simply by subtracting the values of counters for higher thresholds from counts of lower thresholds.

In a further embodiment separate probability density functions are estimated for slow varying effects and fast varying effects. For example, a typical high frequency radio link will suffer from fading which occurs on a time-scale of tens of minutes as well as from noise occurring on a time scale of seconds. It is possible to determine a pdf which characterises the fading and another pdf which characterises the fast variation.

The long time scale variation is essentially deterministic and can be used to predict future channel characteristics. The short time scale variations can be estimated in the way described above and reacted to thereby optimising communications.

A typical problem is shown in FIG. 7 for HF communications. The error-free transmission rate is shown at 35 as a function of time. As can be seen, the transmission channel is suffering from fading. There is also shown at 36 the effects of noise, which occurs on a much faster time scale. The optimum transmission parameters of the communication system are represented by 37.

An embodiment which can achieve the operating mode represented by 37 is shown in FIG. 8. As before the transmitter 1 transmits signals in the form of symbols to a receiver 2. The received signals are analyzed and two or more probability density function histograms are accumulated in pdf means 38 and 39. Each pdf means has a different time constant and therefore represents different time varying interferences. There is a memory means 42 which contains a library of known probability density functions. Comparison means 40 and 42 compare the measured histograms from the pdf means 37 and 38 to those stored in the ^' memory means 42. The comparison means produces one or more indices which identify the library pdfs most closely matching the measured histograms. The indices are fed back as before and used to adjust the transmission parameters of the system.

In an alternative embodiment, instead of forming histograms of the decision variables, the decision variables are compared with each of the library pdf's on a continuous basis. Referring to FIG. 9 curves 43, 44 and 45 represent a library of possible probability density functions (or models). Each measurement 46 is compared with each model and the likelihood that the measurement matches each model is calculated. One measurement may be sufficient to uniquely select a given model, however a number of measurements will normally be necessary. The model corresponding to the maximum likelihood is selected as the appropriate probability density function to use for optimising the transmission.

In a still further embodiment, the invention is incorporated as an improvement to any communication link subject to fading, noise or interference using modems. The feedback control strategy in this embodiment is based upon adapting the error control code rate to maximise the data throughput while maintaining a ceiling on the error probability of the transmission as seen by the user. The optimum error control code rate is determined at the receiver and an index is fed-back to the transmitter which specifies the next error control code rate. The maximum modem transmission rate is used at all times.

Referring now to FIG 10, the data to be transmitted is fed to a buffer 50. An adaptive coder 51 accepts a number of data bits from the buffer 50 depending on the code rate fed-back from the processor 54. The length of the coded data packet in this embodiment is fixed but the number of data bits transmitted in each packet is varied. The rest of the coded data packet is filled with parity bits to allow error correction. The packets are modulated and transmitted by standard modem equipment 52.

The received data is demodulated in a standard modem 53, decoded by an adaptive decoder 55 complimentary to that at the transmitter and stored in a buffer 56 before being output as a continuous data stream. The processor 54 selects the hew code rate which is fed-back to the adaptive coder 51. The processor has two sub-components. The first is a noise pdf estimator which estimates the noise pdf by any of the methods described above. The second component predicts the fading which distorts the signal. The noise pdf and the fading prediction are used to predict the likely errors. The highest rate code is chosen which results in an acceptable post-decoder error probability.

A delay 57 is necessary to ensure that the adaptive decoder does not implement the change before the adaptive coder has received the fed-back index and implemented any change.

Table 1 lists the code rates selectable in this embodiment. Only three bits need be fed-back to uniquely select any one of the eight possible code rates. If fading is severe a low or zero code rate is chosen to maintain an acceptable quality of service (ie. acceptable error rate). For image transmission an acceptable error rate might be 10"⁴ whereas for voice an error rate of 10^"2 is tolerable.

TABLE 1

FIG 11 shows a comparison of the data throughput of a standard modem operating over a fading HF radio channel at two fixed code rates (A and B), utilizing two hybrid automatic-repeat-request schemes (H-ARQ) (C andD) and the adaptive code rate technique of the embodiment (E). The invention clearly results in an increased data troughput of a standard modem compared to existing data transmission quality control methods. This embodiment has been described in further detail including, the theoretical basis, in 'Performance of an Adaptive Code Rate Strategy for Multiplicative Fading Channels', Proceedings of Communications 92 Conference, 20-22 Oct 1992, Sydney, Australia and 'An Adaptive Feedback Communications Scheme for Parallel-tone Modems Operating in Fading Channels', Proceedings of ISSSE Conference, Sep 1992, Paris, France.

The invention herein described provides a method and apparatus whereby the full advantage for communications optimisation may be made of the measured probability density function. By feeding back an index to known probability density functions or transmission state parameters stored in a library only a low bandwidth feedback channel is required. Feeding back this index is equivalent to feeding back the entire probability density function which offers considerable advantages over existing methods of feedback optimisation in that it permits the selection of the optimum transmission parameters for achieving the desired communication task.

The invention is not directed to any particular method of communications optimisation and it will be appreciated by those skilled in the field that once the pdf is determined any error minimisation code may be utilised to determine the transmission state parameters to optimise the communications.

Throughout this specification the purpose has been to illustrate the invention and not to limit this.

Claims

1. A method of optimising the performance of a communication system comprising the steps of : receiving signals at a receiving means; forming decision variables from each received signal; comparing over time the decision variables with values derived from a plurality of models maintained in a library; calculating a likelihood value for each library model said likelihood value being a measure of the degree to which the model agrees with the measured decision variables; selecting the model with the maximum likelihood value; utilizing the selected model to establish criteria for modifying a transmission parameter state of a transmitting means; sending back to the transmitting means an index uniquely specifying the next transmission parameter state; receiving the index at the transmitting means and using the index to select the next parameter state for the transmitting means.

2. The method of claim 1 in which the model with the maximum likelihood value is selected by determining the logarithm of each of the likelihood values, comparing the calculated logarithm of the likelihood value for each library model and choosing the model associated with the largest log-likelihood value.

3. The method of claim 1 wherein the index identifies a possible transmission parameter state from a library of possible transmission parameter states and wherein the library of possible transmission parameter states is contained in a memory means.

4. The method of claim 3 wherein there are identical libraries contained in memory means provided at the transmitting means and at the receiving means.

5. The method of claim 1 wherein the criteria for modifying the transmission parameter state is established by analysing the communication system requirements and a probability density function of the decision variables.

6. A method of assessing and optimising a communication system including the steps of : measuring at least one probability density function of transmission through an associated communication link or channel; calculating from the measured probability density function transmission state parameters required to optimise the communication; transmitting the transmission state parameters from the receiver to the transmitter; and modifying transmission parameters of the transmitter to accord with the transmission state parameters.

7. ^■ A method of assessing and optimising a communication system including the steps of : measuring at least one probability density function of transmission through an associated communication link or channel; calculating from the measured probability density function transmission state parameters required to optimise the communication; comparing the transmission state parameters with a library of transmission state parameters and selecting an index associated with the library transmission state parameters most nearly matching the calculated transmission state parameters; transmitting the index from the receiver to the transmitter; and modifying transmission parameters of the transmitter according to criteria established with the index.

8. A method of assessing and optimising a communication system including the steps of : measuring at least one probability density function of transmission through an associated communication link or channel; comparing the measured probability density function with a library of probability density functions and selecting an index associated with the library probability density function most nearly matching the measured probability density function; transmitting the index from the receiver to the transmitter; and modifying transmission parameters of the transmitter according to criteria established with the index.

9. The method of claim 6, 7 or 8 wherein the step of measuring at least one probability density function is characterised by recording as a histogram value the occurrences or proportion of occurrences over a number of symbol transmissions that the decision variable is categorised as being a particular category.

10. The method of claim 8 wherein the index identifies the location of a probability density function in a first memory means associated with the receiver and in a second memory means associated with the transmitter.

11. The method of claim 8 wherein the step of comparing the measured probability density function with a library of probability density functions is further characterised by the library probability density functions being for the same number of categories as the measured probability density function.

12. An apparatus for optimising digital communication systems including: a bi-directional communication link adapted to transmit and receive a plurality of signals in the form of symbols; measuring means adapted to measure a probability density function associated with the link; memory means containing a library of known probability density functions; comparing means adapted to compare the measured probability density function with the library of known probability density functions and produce an index to the library probability density function which most closely matches the measured probability density function; feed-back means adapted to feed back the index; analysis means adapted to extract from a second memory means the probability density function indicated by the index and to determine appropriate transmission parameters to optimise the digital communication system.

13. An apparatus for optimising digital communication systems including: a bi-directional communication link adapted to transmit and receive a plurality of signals in the form of symbols; measuring means adapted to form decision variables from each received signal; memory means containing a library of known probability density functions; comparing means adapted to compare the measured decision variable with values derived from the library of known probability density functions and produce a likelihood value, said likelihood value being a measure of the degree to which the known probability density function agrees with the measured decision variables and to select the probability density function associated with the maximum likelihood value; calculating means adapted to utilize the selected probability density function to calculate transmission state parameters required to optimise the communication system and to produce an index to a library of transmission state parameters; feed-back means adapted to feed back the index to the transmitter; memory means at the transmitter containing a library of transmission state parameters; and transmitter control means adapted to receive the index, select the transmission ^• state parameters from the memory means according to the index and apply the ^' parameters to the transmitter to optimise the communication system.

14. An apparatus for use in a digital communication system including: first transmitting means adapted to transmit via channel means signals in the form of symbols; first receiving means adapted to receive said signals; demodulator means adapted to form at least one decision variable from a received signal, each decision variable being characterised by a decision attribute which is useful in determining which of a plurality of possible symbols is being received; first memory means containing a library of probability density functions; calculating means adapted to produce an index; second transmission means adapted to transmit an index signal corresponding to said index; second receiver means associated with the first transmitter means and adapted to receive the transmitted index signal; second memory means containing a library of transmission state parameters; and transmitter control means adapted to extract transmission state parameters from the second memory means, according to the index signal, and to control the transmitter to transmit according to the selected transmitter state parameters to optimise the digital communication system.

15. The apparatus of claim 14 wherein the calculating means is adapted to: form at least one measured probability density function of the value of the decision attribute of an associated decision variable in histogram form by repetitively categorising the associated decision variable into categories dependent upon the value of the decision attribute, where the categories . consecutively divide the range of values that the decision attribute can take, and each category has an associated range being a sub-range of the range of values that the decision attribute can take; record for each category the number of times or proportion of times that the value of the decision attribute is within the associated range of the category or the value of the decision variable is within or exceeds the region of the category; compare the measured probability density function with the probability density functions contained in the first memory means; and to provide an index associated with the library probability density function which most closely matches the measured probability density function.

16. The apparatus of claim 14 wherein the calculating means is adapted to compare over time the decision variable with values derived from each of the library probability density functions contained in the first memory means; calculate a likelihood value for each library probability density function, said likelihood value being a measure of the degree to which the value derived from the library probability density function agrees with the measured decision variables; and provide an index associated with the library probability density function having the maximum likelihood value.

17. The apparatus of claim 15 or claim 16 wherein the calculating means is further adapted to categorise the decision variables at the same time that a demodulator demodulates the decision variable into the symbol being received.

16. A method of optimising time-varying communication systems comprising the steps of : repetitively measuring a decision variable; building a first histogram of decision variable measurements using a slow time constant and building a second histogram of decision variables using a fast time constant; comparing the histograms with a library of histograms and selecting respective indices of the histograms which most closely match the measured histograms; feeding back the selected indices; and adjusting transmission parameters of the system in accordance with criteria established with the indices.

17. An apparatus as herein described with reference to the attached drawings.

18. A method as herein described with reference to the attached drawings.