To achieve a better Eb/N0 (bit energy to noise ratio) than todays Rake receiver, a better estimation of the radio channel is needed. This paper proposes to reach this goal with artificial neural networks, which are used for detecting all multipaths in a direct sequence spread spectrum signal.
The output of the adaptive neual net is an estimation of the channel response. To increase the Eb/N0 of direct sequence spread spectrum signals, Rake receivers consist of up to 6 demodulators, which demodulate the strongest few echoes of the signal. Their output signals are collected. The result is a higher stability against noise.
The presented work is also based on using not only one path for reconstructing the transmitted bit stream but nearly all paths. The received PN (pseudo noise) radio signal is sampled with a rate of about four times the chip rate of the PN signal and passed through a despreader that correlates it with the original PN sequence.
In contrast to a conventional receiver, the correlation is made for every delay, realized in a correlator bank whose output is given to a neural network. As neural net, a perceptron with backpropagation is used, with the task to detect all the multipaths in the correlated signal. With conventional filters the problem of detecting only the multipath peaks and not the peaks due to crosscorrelation of a PN sequence with an inverted PN sequence - occuring whenever a bit change from 1 to -1 or inverse - is nearly impossible to handle. The output of the net is the estimated response of the actual channel. With this information containing the delay and the attenuation of each propagation path, the transmitted bit stream can easily be rebuilt. .