Adaptive modem for varying communication channel

Claims

We claim

1. An adaptive modem system for a communication channel exhibiting multipath dispersion, which has a channel response characterized by an anticipated multipath delay time length, comprising:

transmitting means for transmitting signals over the communication channel representing a packet of data made up of a plurality of frames in series, wherein each frame is composed of two adjacent training blocks in series with a data sequence of unknown symbols, which symbols are unknown as containing data to be received and which symbols become known only when received and decoded with respect to an accurate estimate of said channel response and wherein a second of the two training blocks is made up of M known symbols, and a first of the two training blocks has a time length longer than the multipath delay time length and separates the second training block from the data sequence; and

receiving means for receiving the transmitted signals, demodulating the data packet therefrom, and processing each frame thereof, including:

(a) source means for storing and generating the sequence of M known symbols;

(b) channel estimate means for correlating the sequence of M known symbols from said source means with the received M known symbols of the second training block and obtaining an estimate of said channel response as would be associated with an anticipated multipath delay time length in each frame of the data packet, wherein said second training block is isolated by the time length of said first training block so as to be substantially unaffected by any multipath dispersion from the data sequence; and

(c) decision means for decoding the unknown symbols of the data sequence in each frame using the estimate of said channel response as obtained.

2. An adaptive modem system according to claim 1 wherein said two training blocks each have the same time length longer than the multipath delay and have the same M symbols.

3. An adaptive modem system according to claim 1, wherein each packet transmitted by said transmitted means has an initial frame containing the two training blocks and a series of subsequent frames each having the two adjacent training blocks following a data sequence of unknown symbols.

4. An adaptive modem system according to claim 1, wherein said receiving means further includes interpolating means for interpolating an estimate of the channel response at a time position corresponding to any selected part of the data sequence based upon the estimates obtained for the second training block for the current frame and an adjacent frame, and wherein said decision means uses said interpolated estimate for deciding the unknown symbols in at least said selected part of the data sequence.

5. An adaptive modem system according to claim 1, wherein said receiving means further includes demultiplexing means for dividing the data sequence into a series of i data blocks of N symbols each, interpolating means for interpolating an estimate of the channel response for each respective data block at its corresponding time position t.sub.i in the data sequence, and wherein said decision means includes a plurality of i decision loops each having selected filtering functions for processing a respective one of said data blocks, respectively, said filtering functions of each decision loop having tap values which are set in accordance with the interpolated channel response estimate for the corresponding data block.

6. An adaptive modem system according to claim 5 adapted to track the communication channel varying at a rate of 2 Hz or more.

7. An adaptive modem system according to claim 5 configured so as to decode data symbols at a bit-error rate of from 0.001 to 0.00001 for an energy/bit to noise ratio of from about 12 to 16 db for HF transmission at a rate of 4800 bps with 1 Hz fading, 5 msec multipath dispersion.

8. An adaptive modem system for receiving signals transmitted over a communication channel in a packet made up of a plurality of frames in series, wherein each frame includes a training sequence of M known symbols in series with a data sequence of unknown symbols which symbols are unknown as containing data to be received and which symbols become known only when received and decoded with respect to an accurate estimate of said channel response, said channel having a channel response characterized by an anticipated multipath delay time length, comprising:

receiving means for receiving the transmitted signals, demodulating the data packet therefrom, and processing each frame thereof, including:

(a) source means for storing and generating the sequence of M known symbols;

(b) channel estimate means for correlating the sequence of M known symbols from said source means to the received M known symbols of the training sequence in each frame of the data packet, in order to derive an estimate of the channel response for a corresponding time position of the training sequence in each frame with said estimated channel response being that which would be associated with an anticipated multipath time delay; and

(c) decision means for decoding the unknown symbols of the data sequence in each frame, said decision means including the interpolating means for interpolating an estimate of the channel response at a time position corresponding to a selected part of the data sequence based upon the estimates obtained for the training sequence for said frame and an adjacent frame, and wherein said decision means uses said interpolated estimate for deciding the unknown symbols in at least said selected part of the data sequence.

9. An adaptive modem system according to claim 8, wherein said receiving means further includes demultiplexing means for dividing the data sequence into a series of i data blocks of N symbols each, interpolating means for interpolating an estimate of the channel response for each respective data block at its corresponding time position t.sub.i in the data sequence, and wherein said decision means includes a plurality of i decision loops each having selected filtering functions for processing a respective one of said data blocks, respectively, said filtering functions of each decision loop having tap values which are set in accordance with the interpolated channel response estimate for the corresponding data block.

10. An adaptive modem system according to claim 9, wherein said training sequence consists of two training blocks each having the same time length longer than the multipath delay and having the same M symbols, and said two training blocks follow a data sequence in each frame divided into at least two data blocks to be processed in at least two decision loops of said decision means frame-by-frame.

11. An adaptive modem system according to claim 10, wherein each data block is processed through its corresponding decision loop in at least two passes, and each decision loop includes a precanceller for precancelling interference from a following data block, a matched filter for combining input signals for the symbols in each frame from said precanceller into enhanced signals, a feedforward filter and a feedback filter for removing intersymbol interference of adjacent symbols in the enhanced signals from said matched filter, and a decision unit for deciding the data symbols filtered by said feedforward and feedback filters.

12. An adaptive modem system according to claim 11,

wherein on the first pass, a preceding training block of a prior adjacent frame is used to initialize the feedback filter of a first decision loop for a first data block of the current frame, a following training block in the current frame is used for the precanceller of a second decision loop for a following, second data block, and decisions made by the decision unit on the first data block are used to initialize the feedback filter of the second decision loop, and

wherein on the second pass, decisions made on the second data block on the first pass are used in the precanceller of the first decision loop, and decisions on the first data block on the second pass are used to initialize the feedback filter for the second decision loop.

13. A method of operating an adaptive modem system for receiving signals transmitted over a communication channel in a packet made up of a plurality of frames in series, wherein each frame includes a training sequence of M known symbols in series with a data sequence of unknown symbols which symbols are unknown as containing data to be received and which symbols become known only when received and decoded with respect to an accurate estimate of said channel response, said channel having a channel response characterized by an anticipated multipath delay time length, comprising the steps of:

receiving the transmitted signals, demodulating the data packet therefrom, and processing each frame thereof, including:

(a) storing and generating the sequence of M known symbols;

(b) correlating the sequence of M known symbols to the received M known symbols of the training sequence in each frame of the data packet, in order to derive an estimate of the channel response for a time position of the training sequence in each frame with said estimated channel response being that which would be associated with an anticipated multipath time delay, and

(c) interpolating an interpolated estimate of the channel response at a time position corresponding to a selected part of the data sequence intermediate to the time positions of the training sequences for a frame and an adjacent frame based upon the estimates obtained for the training sequences, and using said interpolated estimate for deciding the unknown symbols in at least said selected part of the data sequence.

14. A method of operating an adaptive modem system according to claim 13, further comprising the steps of dividing the data sequence into a series of i data blocks of N symbols each, interpolating an estimate of the channel response for each respective data block at its corresponding time position t.sub.i in the data sequence, and using the interpolated estimates for processing the respective data blocks through a plurality of i decision loops corresponding to said data blocks, respectively, wherein each of the decision loops have selected filtering functions in which respective tap values are set in accordance with the interpolated channel response estimate for the corresponding data block.

15. A method of operating an adaptive modem system for a communication channel exhibiting multipath dispersion, which has a channel response characterized by an anticipated multipath delay time length, comprising the step of:

transmitting signals over the communication channel representing a packet of data made up of a plurality of frames in series, wherein each frame is composed of two adjacent training blocks in series with a data sequence of unknown symbols which symbols are unknown as containing data to be received and which symbols become known only when received and decoded with respect to an accurate estimate of said channel response, and wherein a second of the two training blocks is made up of M known symbols, and a first of the two training blocks has a time length longer than the multipath delay time length and separates the second training block from the data sequence such that an estimate of the channel response for each frame can be obtained by correlating the received signals for the second training block of each frame with a stored version of the M known symbols without interference from the unknown symbols of the data sequence.

16. A method of operating an adaptive modem system according to claim 15, wherein said two training blocks each have the same time length longer than the multipath delay and have the same M symbols.

17. A method of operating an adaptive modem system according to claim 16, wherein said time length of each of said training blocks is longer than about 5 milliseconds for the anticipated multipath delay.

18. A method of operating an adaptive modem system according to claim 15, wherein each packet transmitted by said transmitted means has an initial frame containing the two training blocks and a series of subsequent frames each having the two adjacent training blocks following a data sequence of unknown symbols.

19. A method of operating an adaptive modem system for receiving signals transmitted over a communication channel in a packet made up of a plurality of frames in series, wherein each frame includes a training sequence of M known symbols in series with a data sequence of unknown symbols which symbols are unknown as containing data to be received and which symbols become known only when received and decoded with respect to an accurate estimate of said channel response, said channel having a channel response characterized by an anticipated multipath delay time length, comprising the steps of:

receiving the transmitted signals, demodulating the data packet therefrom, and processing each frame thereof, including:

dividing the data sequence of each frame into a series of at least two data blocks of N symbols each;

processing each of said at least two data blocks through a corresponding one of at least two decision loops in at least two passes, wherein each decision loop includes a matched filter for combining input signals for the symbols in each frame into enhanced signals, a feedforward filter and a feedback filter for removing intersymbol interference of adjacent symbols in the enhanced signals from said matched filter, and a decision unit for deciding the data symbols filtered by said feedforward and feedback filters,

wherein on the first pass, a preceding training sequence of a prior adjacent frame is used to initialize the feedback filter of a first decision loop for a first data block of the current frame, and decisions made by the decision unit on the first data block are used to initialize the feedback filter of a second decision loop for a following, second data block, and

wherein on the second pass, decisions on the first data block on the second pass are used to initialize the feedback filter for the second decision loop.

20. A method of operating an adaptive modem system according to claim 19, wherein each decision loop for the corresponding data block further includes a precanceller for precancelling interference from a following data block from the signals provided to said matched filter, and wherein decisions made on the second data block on the first pass are used in the precanceller of the first decision loop, and a following training sequence in the current frame is used for the precanceller of the second decision loop.

21. A method of operating an adaptive modem system according to claim 19, further comprising the steps of:

(a) storing and generating the sequence of M known symbols;

(b) correlating the sequence of M known symbols to the received M known symbols of the training sequence in each frame of the data packet, in order to derive an estimate of the channel response for a time position of the training sequence in each frame with said estimate of the channel response being that which would be associated with an anticipated multipath time delay; and

(c) interpolating an interpolated estimate of the channel response at a time position corresponding to each data block of the data sequence intermediate the time positions of the training sequences for a frame and an adjacent frame based upon the estimates obtained for the training sequences, and using said interpolated estimate for each data block to set respective tap values of the filtering functions in each corresponding decision loop.

22. A method of operating an adaptive modem system according to claim 19, wherein said training sequence consists of two training blocks each having the same time length longer than the multipath delay and having the same M symbols.

This invention relates to a modem which adapts to a varying communication channel by accurate and ongoing estimation of the channel response in order to decode data signals transmitted over the channel with a high degree of accuracy. In particular, the invention is directed to an adaptive modem for receiving a high frequency transmission in the form of packets made up of alternate sequences of known training symbols and unknown data symbols. The invention uses the known training sequences to estimate the channel response at the instant of the training sequence. The invention then uses the estimated channel response to interpolate the channel response across the unknown data sequence, and divides that data sequence into data blocks for decision processing using interpolated channel estimates.

BACKGROUND OF INVENTION

When data signals are transmitted over a communication path, various types of distortion and noise are introduced. This is due to interference and the changing and dispersive nature of the communication path. A principal form of distortion is the multipath dispersion which occurs when signals propagate along different or reflected paths through a transmission medium to the receiving destination. For example, in high frequency (2 to 30 MHz) transmissions which are bounced off the ionosphere, multipath dispersion is introduced in the transmission in the form of echoes, time delays, fading, phase changes, and other adverse influences of the communication channel. Accordingly, the signals received are not the same as the original message, and when they are demodulated and decoded there are often errors in the output data. Other adverse effects include interference between the transmitted data symbols (intersymbol interference or ISI), and noise which reduces the signal to noise ratio (SNR) at the receiving end.

Various approaches have been developed to compensate for the adverse effects of the communication channel on data transmissions. One approach, such as disclosed in U.S. Pat. No. 4,058,713 to Di Toro, has been to transmit alternating bursts of a known test signal with segments of the original (unknown) data, and to use the known test signal at the receiver to derive an estimate of the channel influence, which estimate is then employed to process the unknown segments of data using a frequency domain data recovery algorithm. With this type of approach, a delay factor is introduced in the format of the transmitted message, which limits the data rate, and detection errors occur if the channel varies significantly at a rate shorter than the period of the data segment. Another method uses channel response estimates to set the coefficients of adaptive equalizers or recursive filters in a time domain data recovery algorithm. A well known technique called "Adaptive Decision Feedback Equalization" (ADFE) has been developed by Bell Laboratory, as described in "Optimum Mean-Square Decision Feedback Equalization", by J. Salz, BTSJ 52, pp. 1341-1373 (1973), and "A Unified Theory of Data-Aided Equalization", by M. S. Mueller and J. Salz, BTSJ 60, pp. 2023-2039 (1981). The performance obtained by the ADFE technique depends very much on the method used to estimate the channel response, because as the channel response changes, the DFE coefficients must be adapted to compensate for the new channel response. In a slowly varying channel, where the fade rate is much less than 1 Hz, e.g. 0.2 Hz, a simple channel tracking algorithm is usually adequate. The adaptation is usually performed within several update cycles in a time period shorter than the fade rate. However, in a more rapidly fading channel, i.e. with fade rates near 1 Hz and above, the update cycles needed to converge on a new channel response often exceed the fade period, and the updated channel response may be outdated before it can be used. The failure to appropriately track the channel response leads to poor performance of the ADFE technique in a fast fading environment. Also, the requirements for digital data transmissions are more stringent for higher data rates of 1200, 2400, 4800 bps or more. Such data rates require more accurate compensation for dynamic channel variations in the range of 1 Hz or higher for high frequency (HF) transmissions.

A faster channel tracking method requires speeding up the update cycles, typically by using a faster microprocessor or by using an improved updating algorithm which requires less computation. As an example of the former approach, reference is made to U.S. Pat. No. 4,365,338 to McRae et al. This patent discloses the transmission of data in packets made up of successive frames, each having a sequence of N known symbols followed by M unknown data symbols. An estimate of the channel response is updated at the receiving end for each frame. A channel tracking algorithm derives an estimate of the channel response by cross correlating an N+M vector of received symbols with the 2N known symbols of the current and previous frames and the channel estimate for the previous frame. The channel estimates are in the form of N+1 weighting coefficients of a transversal filter function applied to the received symbols. The M unknown symbols in each frame are decoded by a "Data Directed Equalization" (DDE) algorithm which calculates the expected error in the decoding decisions on the unknown symbols, and reiterates the error calculation using refined decoding decisions until final decisions on the M unknown symbols are reached having an acceptably low error factor. However, this type of system has the disadvantage that a heavy load of computation is needed to process and decode the data, requiring a specially designed fast array processor, and only achieves acceptable accuracy by extensive iterative recalculation. A major problem with this approach is the inclusion of the unknown symbols in the cross correlation to solve for the channel estimates. The method estimates the channel response based upon unknown inputs from the transmitter and is limited in accuracy and response time as it requires several frames of data to be iteratively processed before the channel estimates converge on true values.

For some applications, such as defense communications, the requirements for transmission and recovery of data are even more stringent. For example, the length of the transmitted data packet may be shortened or changed, and the transmissions may hop in stepped sequence over different frequencies in order to deter interception. Frequency hopping poses severe requirements on channel tracking, since the multipath dispersion is constantly changing over time and is different from one channel to another. For higher data transmission rates, the heavy computational load of the above-mentioned McRae type of system increases its cost and complexity, deteriorates its real time response, and requires trade-offs in the bit-error rate. The computational load can be lowered using an improved data recovery algorithm, for example, as disclosed in U.S. patent application Ser. No. 694,549 filed Jan. 24, 1985, now U.S. Pat. No. 4,761,796 and entitled "High Frequency Spread Spectrum Communication System Terminal", by J. G. Dunn et al. The Dunn system employs an optimal polyphase code (known symbol) sequence at the beginning of each data packet. The code sequence is matched to the anticipated HF channel characteristics in that it employs two repetitions of the polyphase code, with each repetition being greater in length than the multipath delay. Therefore the second repetition of the code does not have any interference from delayed versions of any preceding unknown data symbols. The modem uses the channel estimates derived from processing the received code sequence to set the tap values for linear canceller, feedback equalizer and matched filter functions in a decoding decision loop. The decision loop provides a non-iterative form of equalization to compensate for the multipath effects of the HF channel. The received unknown data is passed twice (or more) through the decision loop to obtain more accurate symbol decisions on the second pass using the preliminary symbol decisions obtained on the first pass. However, the Dunn system has difficulty tracking the channel at channel fade rates above 1 Hz. This is because the channel response is changing rapidly at these higher fade rates: the channel response is well known for the portion or the unknown data symbols immediately following the training symbols, allowing these data symbols to be demodulated with few bit errors; the channel response typically changes significantly by the end of the packet, resulting in many bit errors when the final data symbols are demodulated.

An extension to the Dunn system (one known sequence and one data sequence per packet) would be to break the packet into shorter frames, and to alternate the known and unknown symbol sequences in each frame so that the framing data would always be located closer to the unknown data, allowing the channel response to be better known during the unknown data. Unfortunately this increases the overhead due to the large amount of training data, so that the available data rate is unacceptably limited.

Another extension to the Dunn system would be to break the packet into shorter frames, and include a header frame using Dunn's training format (2 repetitions of a polyphase code, with each repetition exceeding the multipath delay) and then in the subsequent data frames include known data equal to the multipath delay, ala McRae. This allows a fast and accurate initial channel estimate with subsequent updating to track the changing channel. Unfortunately if low-computational cost (conventional least mean square) updating techniques are used, then this approach can get "lost" midway through the packet if one or the frames fades deeply resulting in many bit errors. Since unknown data is included in the updates of the channel estimate, it is not possible for the tracking mechanism to recover unless expensive computational approaches (ala McRae) are used.

SUMMARY OF INVENTION

In view of the above problems and disadvantages of conventional modem systems, a modem system is needed which can provide less computationally intensive data recovery, and can adapt to channel variations higher than 1 Hz, by rapidly deriving accurate channel estimates without excessive training data overhead.

Therefore, it is an object of the present invention to provide a modem system which can adapt to multipath dispersion in a communication channel and to channel variations at high fade rates of 2 Hz or more. It is also desired that the modem system be adapted for data transmissions at higher data rates of 4800 bps or more (i.e. have low training overhead), and provide accurate decoding of the received data signals with very low bit-error rates.

It is a further object of the invention to provide a packet format and channel tracking algorithm which allow accurate channel estimates to be derived quickly from frame to frame, as well as within each frame of data, without heavy computational processing.

Still a further object of the invention is to provide an improved data recovery algorithm which is not computationally intensive and which decodes the data with very low bit error rates, and where the energy/bit to noise ratio may be, low and the transmissions may hop from one frequency to another with each transmitted data packet.

In accordance with the invention, a transmitted data packet is formatted into successive frames. The initial frame carries two identical training sequences. Each succeeding frame is comprised of two identical training sequences, one after another, following a sequence of data. Each training sequence is made up of M known symbols and has a time length longer than the expected rime length of the multipath dispersion in the communication channel. Therefore, the second training sequence of each current frame is isolated from any multipath interference from the preceding unknown data sequence, and a "clean" received signal of the training sequence is obtained for each frame. The "clean" received training sequence enables the receiving modem system to derive very accurate channel estimates for each frame by autocorrelation to the known symbol sequence stored at the receiver. By interpolating across the channel estimates for the current and previous frames, the modem system can track the communication channel by interpolated estimates of the data sequence. Interpolation reduces the amount of training data required to get accurate channel estimates; i.e. interpolation reduces training overhead.

The invention further encompasses an improved data recovery algorithm which divides the received data sequence in each frame into a series of two or more data blocks. Each data block is processed through a respective decision loop using interpolated channel estimates corresponding to the time position of the data block in the data sequence. The interpolated channel estimate for each data block is used to set the tap values for the filter functions of the corresponding decision loop. Each decision loop includes: a precanceller filter function which precancels the expected ISI influence of the following symbols; a matched filter function to combine the signals dispersed by the multipath channel into a reinforced signal having a higher SNR; a feedforward filter function to remove the ISI influence of the following symbols; and a feedback filter function to remove the ISI influence of the preceding symbols introduced as a result of passing the received signals through the matched filter. Each data block is passed through its corresponding decision loop on two or more passes. The unknown symbols for each data block are preliminarily decided in the first pass through the corresponding decision loop. On the second (or subsequent) pass, the first (or prior) pass decisions for each preceding data block are used in the feedback filter function for the following data block, and the decisions for each following data block are used in the precanceller filter function for the preceding data block. The decisions for each data block are thus improved with each additional pass as the prior decisions for the preceding and/or following data blocks are employed in the precanceller or feedback filter functions for the current pass.

The present invention provides at least the following advantages. An accurate channel estimate is obtained for each frame based upon a received known symbol training sequence which does not include any multipath interference from unknown data symbols. The modem system can adapt frame by frame to rapidly varyi