GB2458111A

GB2458111A - OFDM-MIMO System

Info

Publication number: GB2458111A
Application number: GB0803939A
Authority: GB
Inventors: Kassem Benzair
Original assignee: WIMAX COMM Ltd
Current assignee: WIMAX COMM Ltd
Priority date: 2008-03-04
Filing date: 2008-03-04
Publication date: 2009-09-09
Also published as: GB0803939D0

Abstract

A MIMO-OFDM system with two transmit and three receive antennas and uses two diversity modes - the Transmit Diversity mode allows a robust radio link against fading and frequency selective channels and increase the communication range. It consist of transmitting the same signal stream on both transmit antennas with a phase delay factor. The Spatial Multiplexing mode allows a high throughput for a given bandwidth compared with normal communication systems using single input and single output antennas. It consist of transmitting two different signal streams in each transmit antenna. The system is an OFDM-based system, as required by the 802.16x standard, where each symbol consists of 1024 parallel sub-channels called also tones. A slot refers to a single OFDM symbol. A block is defined to be one-third of the active data tones of an OFDM symbol. On the uplink, each of the three blocks can be assigned to a different user. These blocks are allocated by the base station (BTU to different users (NAU) on both the uplink and the downlink. To each block is associated a coding mode, which determines the overall rate of data transfer. Six modes are used in the system, which are chosen depending on a Link Adaptation procedure.

Description

Efficient Implementation of high performance OFDM-MIMO Avido System.

Description:

The AvidotM system is a point.to-multipoint mobile broadband wireless access system that uses MIMO-OFDM technology. MIMO stands for "Multiple-Input, Multiple-Output", which refers to the use of multiple transmit and multiple receive antennas in the Avido system.

OFDM stands for Orthogonal Frequency Division Multiplexing, a method of transmission using a large number of frequency sub carriers (also known as tones) to carry the information.

The AvidotM physical layer has several types of diversity that combine to make the system extremely reliable in the presence of fading and multipath. The use of multiple antennas : *.. yields spatial diversity. The use of OFDM enables us to take advantage of frequency * diversity by introducing an error-correcting code that is spread across numerous tones. If a data unit is received in error, a request is relayed back to the transmitter to retransmit the erroneous data. This use of retransmission at the MAC layer provides a sort of temporal diversity in the case of extended fading. Higher data rates can be achieved through spatial multiplexing.

Transmission and reception procedures On the downlink, there are 2 transmit antennas and 3 receive antennas. On the upliak, there is 1 transmit antenna and there are 3 receive antennas.

To take advantage of the 2 transmit antennas on the downlink, there are two schemes being considered: Spatial multiplexing and transmit diversity. The transmit and receive schemes are summarized in the table listed below.

Scheme Antennas Transmit Receive Uplink itx, 3 rx -MRC for 1x3 channel Downlink (sp.mux) 2 tx, 3 rx Spatial multiplex MMSE linear equalizer For 2x3 channel Downlink (tx. div) 1 sig. 2 tx, 3 rx transmit diversity MRC for 1x3 channel

_________________________ * * S...

: MMSE stands for Minimum Mean Square Error, and MRC stands for Maximum Ratio Combining. * S. * . . *5IS

*....: On the transmit side, in spatial multiplexing, two separate streams are sent across the two * antennas. These two streams are coded separately using one of the 6 downlink coding modes. On the other hand, for transmit diversity, a single stream is coded and then the output signal is transmitted across both antennas. The simplest form of transmit diversity is to send a delayed version of the signal. For OFOM systems, it is appropriate to use circular diversity, where a circularly delayed version of the signal is sent (see figure 1).

Hence the circular delay is applied uniformly to all signals that are transmitted on the second antenna, including training tones and spatial multiplexed data.

Also, the use of "space-time" codes, more appropriately called "space-frequency" codes, in this case is considered for encoding a single stream of data across both antennas. This approach represents a combination of the spatial multiplexing and the transmit diversity schemes.

Training is performed using training tones based on a pseudo-Random Binary Sequence (PRBS). In the uplink, the 3x1 channel coefficients are estimated from the training tones embedded in every data slot. In the downlink, the 3x2 channel coefficients are estimated from the training data in a specific Channel Identification slot. Training tones are provided for each of the two transmit antennas in the downlink, so that all coefficients of the channel matrix are estimated separately. Using these estimated channel coefficients, an interpolation filter is used to find the channel coefficients for tones located between the training tones.

On the receive side, there are two types of receivers, depending on the number of distinct transmitted signals.

In the case of the downlink using spatial multiplexing, an MMSE receiver is used to determine the two transmitted signals from the 3 received signals. The solution for the MMSE receiver is given by: X = H°(MH' +E)1Y where V is the vector of 3 received signals, and is the 3x3 covariance matrix of the channel noise. On the downlink in spatial multiplex mode, H is a 3x2 channel matrix, and the MMSE receiver filter F = H°(HH + �)1 isa 2x3 matrix. * ** * * * * ** **** * S

If there is an interfering source, it is possible to suppress this interfering signal by measuring the covarlance matrix 1 of the noise. Then using the MMSE filter F= H(HH +)1 *.. allows the receiver to suppress this interference. S..

S

* * In the case that there is no interference, it is assumed that the noises on the different antennas are independent and have equal energies, so that = 0213, the MMSE receiver *5S*SS * . can also be written as: F=H(HH +2I).1 = (H°H+a212)1H' In the case of the uplink, there is only one antenna, so that the channel matrix has dimensions 3x1. Also, in the case of the downlink in transmit diversity mode, there is only 1 signal to estimate, so that the 3x2 channel matrix (found using the channel ID slot tones, following by interpolation) can be collapsed into a 3x1 matrix by adding the columns. (Since the training tones on the second antenna are also affected by the circular delay transmit diversity scheme, there is no phase factor needed in this computation).

H11 2 Jill +Jf H32 = H21 H22 H31 = H21 + "31 32 "31 + 1132 Hence, for both uplink, and downlink in transmit diversity mode, it is possible to write H as a 3x1 matrix.

Then to calculate the receive filter F, we use maximal ratio combining (MRC), in which the receive filter is a normalized matched filter, so that: F = (HH)1H where F is a 1x3 matrix.

Converting QAM signals to soft metrics The result of applying the filter F is then either 1 complex signal or 2 complex signals, corresponding to an estimate of the original constellation point from a single transmit antenna. To convert this signal into soft information for the Viterbi decoder, it is necessary to find soft metrics for each bit from the complex received signal. * ** * S S

Suppose that the bits in the Gray code labeling are denoted by X1, ,.. ., X3, where S is *15* the number of bits per QAM. For each bit, the goal is to calculate a metric based on the log-likelihood ratio of the probability. If the metric is positive, the bit is more likely to be a 1. If negative, it is more likely to be 0. The magnitude of the metric indicates the degree of certainty about the decision on the bit. * ** * S S ***.

Let V indicate the received complex signal. If there is no noise and the channel has been equalized perfectly, then V should be equal to one of the constellation points, making the estimation of the bits easy. If V is located in between several constellation points, then it is possible to make estimates of the probability (or equivalently, log-likelihood ratios) for each bit in the Gray-code labelling.

LLR(X.)=log" =i) p(x=oiv) -XLt.X,=I -log XLLXgO P(Yix) -.% St X1 -og p(vx) x LL X,0 where X represents both the bit labeling as well as the constellation point, depending on the context. For the numerator and denominator, the summation is over all constellation points where the i-th bit of X is fixed to be 1 or 0, respectively. Bayes rule is used to reverse the conditioning, where it is assumed that the prior information on the constellation points is uniform. Then the noise, which includes noise-enhancement due to the equalizer, is assumed to be Gaussian with variance.,,2 Then the probability of receiving V when the constellation point X was transmitted is I x) = exp1-(Y -x)2 Jcr11 t 2a11 By making the assumption that we only consider the closest points in both the numerator and the denominator, C0 = arg - C1 = -this expression simplifies as follows: * ** **** * I 1 **.* 1 2 exp-211Y-C111 :r LLR(Xj= log 2o exp -2 IP'C011 * ::::: = 2H2 (2Y(c, -c0)+11c01f2 r1 112) 11.1.. * *

This metric serves as the input to the Viterbi decoder for the convolutional code. It should be noted that this expression includes a factor corresponding to the inverse of the noise variance.

In the case of 4QAM, this expression simplifies greatly because all the constellation points have the same magnitude -.--. Hence, LLR(X1) = a,' LLR(X2)= _.!-r(JiIm(Y)) 0,' Weighting the soft metrics The use of a linear equalizer can lead to an enhancement of the channel noise, so that it is necessary to take the channel equalizer into account when estimating the noise variance.

This adjustment allows the convolutional decoder to weight the information it receives from different tones and different antennas. In this way, the information from tones or antennas that have bad channel conditions are decreased in importance by the weighting, while the information from good tones and antennas is boosted by the weighting. This allows for better performance under frequency-varying conditions and also when the channel matrix is poorly conditioned. In a sense, this use of weighting and coding can accomplish a similar purpose as transmit optimization and bit-loading.

It is assumed that the actual channel noise is fiat and Gaussian (with variance a2), so that it is the same regardless of the frequency tones in the OFDM system. In addition, it is assumed that there is no correlation between the noise on different receive antennas, so that the noise covariance matrix is a213. It should be noted that by allowing for more complicated noise covariance matrices, It is possible to perform interference cancellation or interference suppression. Then the variation of the channel matrix over different frequency tones (due to delay spread, fading and also the transmit diversity scheme) results in different noise variances for the different tones as well as for the different antennas. S..

S S..

S

* In particular, consider the filter F = H'(HH + E)1, as in the MRC or MMSE. In the case of 1 transmit signal, the noise is also enhanced by the filter F, so that the resulting noise : covariance matrix for the tone is FF'. In addition, since the filter is MMSE and not zero- * forcing, it does not exactly invert the channel, so that it is necessary to include this equalization noise, if V = HX + N, the equalizer noise takes the form of: FV-X=(FH-I)X4-FN where FH I may be non-zero because the MMSE linear equalizer does not perfectly cancel out the interfering signal. Hence, the noise covariance matrix is given by = (FH -IXFH -i) EXI2)+ FF which also happens to be the expression for the mean-squared error.

The transmitted signals are assumed to be independent and normalized to have power 1, so that EXI2) is the identity matrix. This expression can then be further simplified by using the definition of F = H(HH + H =(FH-IXFH-Ii +FEF =FHHF -2H'(HH' �1H+l+FF° =F(HH +)-2H(HH �E)'H+I =I-H!IH �1H Then the noise variance for input to the QAM to soft metric function for each antenna is given by taking the diagonal elements of the matrix Z11.

When there are 2 transmitted signals, the equalized covariance matrix is: 2 H*(HH* + which is a 2x2 matrix. Then for the first antenna, the appropriate noise variance to use is (ZH), and for the second antenna, the appropriate noise variance is (ER). Depending on the conditioning of the channel matrix, these two numbers can be very different. This allows the Viterbi decoder to weight the estimates for the two transmit antennas * .* appropriately. In the case of independent noise (with no interfering source), we have E = 2j3,and this expression can be simplified as follows: E, =I.Ho(HH*+ozI)IH I-(HH+o21)1HH : =o2(HH+o21)1 S...

Since c is assumed to be uniform over the entire slot, it is a common constant that can be ignored without affecting the performance of the Viterbi decoder. The remaining of this expression (H5H + .2)4 is known as the Noise-Enhancement Factor (NEF). This is 2x2 matrix, whose diagonal entries are the NEF5 for the two antennas. The reciprocals i/diag((H'H �cr2lY1) is known as the Reciprocal Noise Enhancement Factor (RNEF). Note that this RNEF will not be equal to the diagonals of HH + cr2I, in general, since there is an additional factor Introduced in the inversion due to the off-diagonal elements.

When there is 1 transmitted signal, a MRC filter is used, so that the noise variance is the scalar = a2 (H*H)l. Thus the NEF is (HH)1, and the RNEF is simply HH. This applies to both the uplink and the case of transmit diversity on the downlink.