-It is known that in the personal communication systems environment, additive noise is not Gaussian. Experimental studies show that this type of noise obeys the class A Middleton man-made noise statistics.The influence of such a noise is very severe on the probability error rate for a matched filter detector (1). To improve the detection performance, a non-linear detector should be used. This structure is based on the maximum likelihood approach which exploits multiple samples of the incoming signal. The improvement has been established by Spaulding and Middleton for the binary case (2). This paper generalizes this approach to M-ary signal detection, namely M-QAM, which is the most popular scheme in practice.
I. INTRODUCTIONDigital communication systems are being used more and more in consumer products, such as Personal Communication Systems (PCS). Their performance is generally defined by the symbol or bit error rate in the presence of additive white Gaussian noise. It is well known that the household environment is subjected to man-made noise (e.g. microwave ovens, computers, television receivers,...). Experimentally, the observed man-made noise appears as large impulses occurring during the detection interval (3), (4). In a series of papers by the authors, it was shown that this type of non-Gaussian noise has a dramatic effect on the performance of QAM modems when linear (matched fitter) detection is used (5), (6), (7). The performance is shown to be strongly dependent on two parameters: the impulsive index (A) and the power ratio of the impulsive noise component and the Gaussian noise component ( ). Recently, the multicarrier modulation (MCM) transmission scheme was shown to be effective in combatting this type of noise by increasing the impulsive index of the noise (7). However, the overall performance of MCM in nonGaussian noise can be improved if optimal detection is used on each of the sub-carriers. (2) proposed the use of multiple samples of received symbols, then applied the maximum likelihood test to construct the optimum receiver. For M-QAM detection, this approach can be used to improve the receiver performance. This paper presents bounds for the symbol error rate of the optimum passband and baseband receivers.
Spaulding and Middleton
II. NOISE MODEL AND MATCHED FILTER RECEIVERIn this work, the man-made noise is modeled using Middleton class A non-Gaussian noise statistics (8). Let N(t) be the noise at the output of the narrowband IF stage. It can be expressed asThe Middleton class A statistics of the noise envelope ρ(t) is given by (2) As suggested by the CCITR (4), the phase of non-Gaussian noise is assumed uniformly distributed over the interval [0, 2π] and statistically independent of its envelope. Using this assumption, the joint probability density function of the in-phase and quadrature noise components at the output of the matched filter receiver is given by