The two models attached to this solution demonstrate how to use the MLSE block with the Multipath Rayleigh Fading Channel block. The models were saved in MATLAB 7.2 (R2006b). They should be simulated in that version of MATLAB.
Here are some of the salient features of the model mmlse_dynamic_coeffs_6b.mdl.
1. In this model, the dynamically evolving channel coefficients are input to
the MLSE Equalizer block. This model shows the MLSE block being used in
a mode that is more in keeping with a typical wireless channel.
2. The sample time of the Bernoulli Binary Generator block is set to be 5e-6, which corresponds to a bit rate of 200 kbps. Although low, this is typical of a wireless data channel.
3. The Multipath Rayleigh Fading Channel (MRFC) block has a maximum Doppler shift of 30 Hz, which is a realistic physical value. If you want to see a more quickly varying channel, you can increase that value.
4. The Delay vector of the MRFC block is [0 5e-4/50], which corresponds to two consecutive sample times of the input data. This can always be changed, but it is the simplest delay vector for a two-path channel.
5. Even though this model simulates a two-path channel, the technique should be effective for n-path channels.
6. The Gain vector of the MRFC block is set to [0 10], which is completely arbitrary. This can be changed to almost anything.
7. The gain vector is normalized to 0 dB, so that the average power input to the AWGN block is 1 W.
8. The subsystem "Format Channel Coefficients for MLSE" includes a Zero-Order Hold block that sets the sample time to be the same as that of the data being input to the MLSE block. This is very important, and without it, the model will break.
9. In the MLSE Equalizer block, the traceback depth is set to 10. This parameter can be played with to determine its effect on performance.
10. A Receive delay of 20 is used in the Error Rate Calculation block. The delay of 20 arises because of a traceback depth of 10 and the fact that the QPSK demod
takes in a frame of Nx1 and outputs a frame of 2Nx1.
Note: These these results are not calibrated with expected theoretical values. The
book Digital Communications, 4th Ed. by Proakis discusses this topic in section 10.1.4.
In the model, mmlse_dynamic_coeffs_6b.mdl, only one channel estimate is sent to the MLSE block per frame of data. So even though the channel changes every sample, the channel estimate is used for the whole frame of 50 samples. The Doppler frequency of the Multipath Rayleigh Fading Channel block is selected to be small enough that it does not make much of a difference. However, if you greatly increase the Doppler frequency to 300 or 3000 Hz, you will see a noticeable degradation in BER performance.
The mmlse_dynamic_coeffs_small_frame.mdl is an improved model that does not suffer from this problem. This model creates frames of size 1 that are input into the MLSE block, and the 2x1 channel estimates that are also input to the block are applicable for that sample only. The downside of this approach is that it is not as efficient as running with larger frames. The upside is that the channel estimates are perfect every time.
