This letter proposes a new time-domain equalizer (TEQ) training algorithm aimed at minimizing the total interblock interference power seen by an orthogonal frequency division multiplexing receiver. The new TEQ achieves this goal by emphasizing the suppression of the outer edges of the channel response more with the help of a weight matrix.

Orthogonal frequency-division multiplexing (OFDM) is a popular line code scheme that has found application in wideband modem technologies such has asynchronous digital subscriber loop, high bit rate digital subscriber loop, in addition to digital audio broadcast, digital television terrestrial broadcast, and mobile data communication. To prevent interblock interference (IBI), OFDM inserts a guard time sequence, called the cyclic prefix (CP), between each transmitted block. In practice, CP is rarely long enough for the complete elimination of IBI. Understanding the nature of IBI is crucial in subduing its adverse effects. This study characterizes the power spectral density of the IBI in terms of the channel impulse response and the cyclic prefix length. One of the key findings is that the samples of the impulse response tails do not contribute to IBI power uniformly, but rather their contributions grow linearly as the samples move closer to the edges of impulse response. Based on this observation, the study proposes a time of reference computation technique that minimizes the total IBI power as seen by the receiver.

We analyze the performance of the time-frequency representation method that utilizes the Gamma filter. Gamma filter which can be implemented as a cascade of identical first order lowpass filters generates at its taps the Poisson moments of the input signal. These moments carry spectral information about the recent history of the input signal. Due to the inherent time window embedded in the Gamma filter, the moments are local both in time and frequency. Hence, they can be used to construct a time-frequency representation as an alternative to the conventional methods of short term Fourier transform (STFT), cepstrum, etc. The appeal of the proposed method comes from the fact that in the analog domain the moments are readily available as a continuous time electrical signal and can be physically measured, rather than computed offline by a digital computer. Furthermore, for a faithful representation, it is sufficient to observe the moments at the information rate (nonstationarity rate) rather than the usually higher Nyquist rate. The observed moments can be fed into an artificial neural network (ANN) for tasks like prediction, classification and identification. This work studies the performance of the proposed representation scheme as a function of the system parameters, such as; time scale, number of moments and number of bands on the estimation quality.

We propose to use the Gamma filter as a continuous time spectral feature extractor for the preprocessing of speech signals. The Gamma filter is a simple analog structure which can be implemented as a cascade of identical first order low-pass filters. The filter generates the Poisson moments of the input signal at its taps. These moments carry spectral information about the recent history of the input signal and in return they can be used to construct a time-frequency representation alternative to the conventional methods of the short-term Fourier transform, cepstrum, etc. The appeal of the proposed method comes from the fact that in the analog domain the Poisson moments are readily available as a continuous time electrical signal and can be physically measured rather than computed offline by a digital computer. With this convenience, the speed of the discrete time processor following the preprocessor is independent of the highest frequency of the input signal, but is constrained by the stationarity interval of the signal. The moments can be directly fed into artificial neural networks (ANNs) for tasks like classification and identification of time-varying signals like speech.

Gamma filters are a special class of generalized feedforward filters where feedbacks are allowed only locally. The authors present the conditions for the selection of optimal parameters which are the weights and the memory depth of the filter. The conditions for these two set of parameters are decoupled from each other. This allows a matched filter implementation which gives an estimate of the memory depth.>

Presents a vector space framework to study short-term memory filters in dynamic neural networks. The authors define parameters to quantify the function of feedforward and recursive linear memory filters. They show, using vector spaces, what is the optimization problem solved by the PEs of the first hidden layer of the single input focused network architecture. Due to the special properties of the gamma bases, recursion brings an extra parameter lambda (the time constant of the leaky integrator) that displaces the memory manifold towards the desired signal when the mean square error is minimized. In contrast, for the feedforward memory filter the angle between the desired signal and the memory manifold is fixed for a given memory order. The adaptation of the feedback parameter can be done using gradient descent, but the optimization is nonconvex.