Postgraduate Conference Abstract



Inter symbol Interference cancellation for a mobile cellular receiver using mixed lower and higher order statistics

Mobile cellular communications must operate in channels, subject to high levels of multi-path interference, resulting in significant Inter Symbol Interference (ISI) in the mobile receiver. If uncorrected, the high levels of ISI can lead to bit errors occurring in the receiver process. In this case, the interference is represented as a Finite Impulse Response (FIR) filter with each tap of the filter representing the phase, delay and amplitude of each multi-path component. ISI may also be caused by transmission channels with limited bandwidth. Such a system would be a Infinite Impulse Response (IIR) filter resulting in a spreading of the digital signal pulses leading adjacent pulses to overlap and hence interfere.

ISI can be of particular concern in a Code Division Multiple Access (CDMA) system. This system requires that multi-user signals are simultaneously encoded with near orthogonal code sequences which are therefore highly uncorrelated. Due to the time delays of the multi-path components, ISI can cause multi-user interference to become highly correlated to the desired signal and therefore impossible to remove using a typical CDMA de-correlating decoder. AS CDMA is a spread spectrum technique, it is resilient to the overall Mean Square Error (MSE) in the system and ISI is therefore by far the most important source of error in a CDMA receiver.

It is therefore desirable to reduce the ISI in a system as far as possible and for this an equalizer is normally employed. The standard Least Mean Squared (LMS) based equalizer is designed to reduce MSE and in low Signal-to-Noise Ratio environments, common in spread spectrum communications, is inefficient at reducing the ISI of a system. Accordingly this paper will present an adaptive equalization technique using higher order statistics, which in theory are unaffected by Gaussian noise. In practice, an exact estimate of the higher order cumulant values cannot be made but if a close enough approximation can be found, then the effect of Gaussian noise will be greatly reduced and hence the ISI performance of the equalizer will not degrade substantially with low SNR values.

The equalizer training algorithm exploits the form of the CDMA signal, to generate a simple higher order cumulant error function. The CDMA system employs orthogonal sets of periodic digital spreading functions. This orthogonality results in a cross-cumulant matrix with significant values only at lags corresponding to multi-user signal spreading code delays. This simple cumulant structure is deformed by ISI and so equalizer training can be achieved by minimizing these deformations.

A hybrid LMS-Higher order version of the above technique has also been developed. This has the joint advantages of reducing MSE whilst retaining a high level of ISI reduction, and preventing the system from converging to a local minima.