CHANNEL ESTIMATION USING EXPECTATION MAXIMISATION FOR SPACE-TIME COMMUNICATIONS SYSTEMS
Field of the Invention
The present invention relates to a wireless communications system employing 5 space-time coding. Such systems are also known as MIMO (multi-in/multi-out) systems because both the transmitter and the receiver have multiple antennas.
Background to the Invention
Mobile phones are developing into powerful multimedia terminals capable of 10 streaming high quality audio and video. Work is in progress for providing data rates of up to 2 Mbps to mobile phones and other personal communications devices.
It has been suggested that the spatial properties of the multipath environments, in which mobile phones, wireless network interface cards and the like are usually used, 15 could be employed to increase bit rates without increasing the bandwidth demand.
The main problem with such systems is that the signals transmitted from all of the transmitter's antennas will generally be received by all of the receiver's antennas. In order to separate out the transmitted bit streams, knowledge of the transfer
20 functions of all of the paths between transmit and receive antennas is required. In a mobile environment in particular, these transfer functions will be continuously changing. One solution to this problem is to include predetermined training and pilot sequences in the transmitted signals which can be used for channel estimation by the receiver. This approach has the disadvantage that the bandwidth of the
25 channel is reduced by the need to include the training and pilot sequences.
Summary of the Invention
The approach, adopted in the present invention, is the use of a generalized Expectation Maximization algorithm.
30
According to the present invention, there is provided a receiver for receiving a space-time coded signal, the receiver comprising a plurality antennas, rf processing means for outputting baseband signal samples in dependence on signals received by
means for outputting baseband signal samples in dependence on signals received by the antennas, signal processing means for performing channel estimation on the basis of said baseband signal samples, characterised in that the signal processing means is configured to employ an Expectation Maximization technique, employing said baseband signal samples and a statistical model of the transmitted signal class for which the receiver is adapted.
Thus, rather than using a true posterior probability in the expectation step, a statistical model of the transmitted signal is used as the basis for an approximation of the posterior probability.
The present invention enables the channel estimation to be performed from the payload symbols and does not require special training and/or pilot symbols.
Preferably, the technique comprises iteratively calculating a channel estimate on the basis of a transmitted signal estimate and calculating a transmitted signal estimate on the basis of a channel estimate. The statistical model may comprise probabilities for each valid transmitted signal state which will depend on the modulation and coding employed. Preferably, the transmitted signal estimate comprises a variational mean and the second order moment of the transmitted signal.
Preferably, the signal processing means has access to a value for the number of transmitting antennas being used to provide the received signal and the channel estimate comprises an estimate of the transfer function of each transmitting antenna to receiving antenna path. More preferably, the channel estimate comprises an estimate of the noise contribution to the received signals.
Preferably, the technique employed uses blocks of samples to produce a channel estimate. More preferably, the signal processing means is configured such that the transmitted signals are recovered from a block of samples using the channel estimate determined therefrom.
According to the present invention, there is also provided a space-time coded communication system comprising a transmitter and a receiver (11), according to the present invention, for receiving transmissions from the transmitter, wherein the transmitter comprises a plurality of antennas and first and second signal sources producing and supplying phase-locked carriers, having the same frequency and modulated synchronously with different data, to respective ones of said antennas.
Preferably, the transmitter is configured to transmit a broadcast control channel signal using only one or both of its antennas to provide a timebase reference to the receiver. In this case, space-time coding is not employed for the broadcast control channel signal.
Brief Description of the Drawings
Figure 1 is a functional block diagram of a transmitting station and a receiving station in a system according to the present invention;
Figure 2 is a flowchart of a channel estimation process according to the present invention; and
Figure 3 is a block diagram of a transceiver according to the present invention.
Detailed Description of the Preferred Embodiment
Referring to Figure 1 , a transmitting station 1 comprises a source 2 of data for transmission, a data stream divider 3 for dividing the stream of data from the source of data 2 into first. and second substreams, first and second odulator/PA (power amplifier) modules 4, 5 for modulating a carrier with respective ones of the substreams and first and second antennas 6, 7 fed from the first and second modulator/PA modules 4, 5. The modulator/PA modules 4, 5 use the same carrier frequency, power. level and modulation type. The modulation of the carriers of the modulator/PA modules 4, 5 is synchronised and the carriers themselves are phase- locked and preferably derived from the same source The antennas 6, 7 are arranged to be at least λ/2 apart or so that them have significantly different, and preferably complementary, radiation patterns.
The transmitting station 1 transmit the substreams in 0.577 ms bursts and uses a block code. Each burst comprises 147 symbol periods, 142 data symbols with 2.5 symbol guard periods at either end. Thus, the duration of one symbol is about 4μs and the length of one symbol in free space is about 1.2 km.
A receiving station 11 comprises first and second antennas 12, 13, first and second front end/IF (intermediate frequency) modules 14, 15 for processing the signals received by the first and second antennas 12, 13 and outputting IQ (in-phase and quadrature) baseband sample pairs, a digital signal processor 16 for extracting the transmitted data from the received signals and a channel estimator 17.
The antennas 12, 13 are arranged to meet that same constraints as those of the transmitting station 1.
The use of complementary pulse shaping filters in the transmitting and receiving stations 1 , 11 means that the output IQ sample pairs will be constant for a given symbol, irrespective of where in the symbol the sample was take. Since the length of a symbol is about 1.2 km, relatively small differences in the lengths of the various paths, such as would be expected in urban and indoor environments, can be disregarded. In other words, it can be assumed that the symbols contributing to any pair of IQ sample pairs, output by the front end/IF modules 14, 15, can be assumed to have been transmitted together and that there is no significant contribution from preceding or succeeding symbols. Any minor contribution from neighbouring symbols is treated as noise. This also means that the receiving station's time base can be set with sufficient accuracy from a broadcast control channel transmitted from only one of the transmitter antennas 6, 7 or simultaneously from both of the transmitter antennas. High data rates are not necessary for broadcast control channels, such as are used in mobile phone systems, and consequently they would not normally need to be transmitted using space-time coding.
The digital signal processor 16 receives the IQ sample pairs from the front end/IF modules 14, 15 and reconstructs the transmitted substreams. The digital signal processor 16 then decodes the substreams and outputs the result. In order to
reconstruct the transmitted substreams, a channel estimation is required. In the present embodiment, the channel estimator 17 generates values for the transmitted symbols in the channel estimation process. Consequently, the channel estimator 17 provides values for the transmitted symbols back to the digital signal processor 16 rather than a channel estimate as such.
The channel estimator 17 comprises a processor programmed to estimate
i.e. the transfer functions of the paths between each of the transmitting station's antennas 6, 7 and each of the receiving station's antennas 12, 13, from the IQ sample pairs of the current burst from the front end/IF modules 14, 15, and
which is the covariance of the noise for the various transmitter and receiver antenna combinations.
The IQ sample pairs output by the front end/IF modules 14, 15 are supplied to the channel estimator 17, one burst's worth at a time. In the channel estimator 17, a received burst R is represented by an array of the form:
where there is column for each symbol position (1 to 142) in a burst and a row for each receiving antenna.
The channel estimator 17 estimates H iteratively using R.
Referring to Figure 2, the channel estimator 17 first assigns initial values to J and φ (step si) according to
where the subscript i-1 indicates the preceding iteration or, in the case of the first iteration, the last value used in the channel estimation for the preceding burst and
f indicates the adjoint. The use of block coding in the transmitters renders J, diagonal.
Once J and φ have been calculated, a parameter γ is calculated for each antenna in accordance with: ,. = ^. -(J; - -f g(J,.)XS)w (4) where (S) is the mean field estimates of the posterior mean of the current burst S (step s2).
Having calculated γ, a parameter λ is obtained for each antenna and each symbol from / according to:
where m is the antenna index and / is the symbol index (step s3). Thus, the values for λ
m/, are taken from the diagonal of J according to the antenna index.
T' and λ are the natural parameters of the best Gaussian fit to the posterior probability of the transmitted symbols.
(S) can now be calculated for each antenna in accordance with:
where is an index of points in the constellation of valid states for the modulation being employed, e.g. 8 for 8PSK, ξk is the complex value of the kth point in the constellation and pk is the prior probability of a symbol at the kth point in the constellation being transmitted (step s4). pk is dependent on the constellation and may be affected by the coding used.
Thereafter, (jSf ) is calculated in accordance with:
(step s5).
( I S j ) is then improved in accordance with:
,|2 . sn i.mproved =χ+(\s\}(s) (g) where
1-7
A x m'm ■ [ rL
A, =diag(A ,...,AMI) and
Λ„, = 2 S m. 'lm. J„
(step s6)
New values for iJ nd Σ can now be calculated according to:
where is the number of symbols in a burst, (steps s7 and s8)
If the new values of i-Tand Σ have not changed insignificantly from the values used in Equations (2) and (3), e.g. at least one element has changed by more than 10%, the process is repeated and the program flow returns to step si . However, if the values of i-f and Σ have not changed, or have changed insignificantly, the current estimate of the transmitted symbols (S) is output to the digital signal processor 16.
The iterative algorithm is well-behaved and will converge on an acceptable solution in a reasonable number of iterations.
The symbols (S) are then decoded in a conventional manner by the digital signal processor 16.
Step s6 is computationally intensive and gives the best results. However, it may be omitted.
If bi-directional communication is desired, a station will require both the transmitting and receiving capabilities described above.
Referring to Figure 3, a transceiver arrangement, suitable for a mobile phone or a wireless network interface card, comprises first and second antennas 21, 22 arranged to meet the requirement set out above, first and second antenna switches 23, 24 coupled to respective antennas, first and second front end/IF modules 25, 26 coupled to respective antenna switches 23, 24, first and second modulator/PA modules 27, 28 coupled to repective antenna switches 23, 24, a digital signal processor 29 which processes input data/speech signals for transmission by the modulator/PA modules 27, 28 and received signals to produce output baseband data/speech signals, and a channel estimator 30.
The digital signal processor 29 and the channel estimator 30 operate in the manner described above.
An indoor environment experienced by a static stations may be much less dynamic than that experienced by a moving mobile phone. Consequently, it may not be necessary to recalculate i-T for every burst.
Thus, in a modification to the receiving processes described above, the channel estimator 17, 30, only performs the iterative channel estimation process when the bit error rate (BER) of the demodulated data, provided by the digital signal processor 16, 29 increases by a threshold amount. Instead the most recent versions of -f and -∑" are used to obtain the transmitted symbols 5 from the baseband received signal R. A poor BER does not of itself indicate the need for a new channel estimation since the propagation paths may be static but lossy.
In the foregoing, the channel estimator 17, 30 has been shown separate from the digital signal processor 16, 29. This need not be the case and the function of the channel estimator 17, 30 may be performed by the digital signal processor 16, 29.
Where direct conversion receiving techniques are employed, the front end outputs the IQ sample pairs directly without the need for any IF processing.
Equations (1) to (10) above should be understood to be directions to the skilled person as to the result to be achieved, rather than being prescriptive of the actual program code required.
The embodiments described above employ two transmitter antennas and two receiver antennas. However, the technique is readily scalable to accommodate larger numbers of transmitter and receiver antennas. The technique can also be applied to assymmetrical systems in which the number of transmitter antennas is different from the number of receiver antennas.