A Nondata-Aided SNR Estimation Technique for Multilevel Modulations Exploiting Signal Cyclostationarity

Abstract

Signal-to-noise ratio (SNR) estimators of linear modulation schemes usually operate at one sample per symbol at the matched filter output. In this paper we propose a new method for estimating the SNR in the complex additive white Gaussian noise (AWGN) channel that operates directly on the oversampled cyclostationary signal at the matched filter input. Exploiting cyclostationarity proves to be advantageous due to the fact that a signal-free Euclidean noise subspace can be identified such that only second order moments of the received waveform need to be computed. The proposed method is nondata-aided (NDA), as well as constellation and phase independent, and only requires prior timing synchronization to fully exploit the cyclostationarity property. The estimator can also be applied to nonconstant modulus constellations without requiring any tuning, which is a feature not found in existing approaches. Implementation aspects and simpler suboptimal solutions are also provided.

Cyclostationarity: New trends and applications

“…Further applications are in the following subjects: capacity evaluation of second-order cyclostationary complex Gaussian noise channels [140], orthogonal overlay channels [371], and power line communication channels [319], timing and other signal parameter estimation [101,134,137,168,224,229,236,246,278,317,334], source separation [16,44,106,169], frequency-domain equalizer (FDE) design [374], blind multiple-input multiple-output (MIMO) system identification [299], signal power (SNR), and signal-to-interference-and-noise ratio (SINR) estimation [8,147,294], Doppler spread estimation [376], microDoppler estimation [214], noise modeling in power lines [130,182,318], and radio frequency interference (RFI) mitigation in radio astronomy [145].…”

Cramér-Rao Bounds for SNR Estimation of Oversampled Linearly Modulated Signals

“…It is assumed that F and " are known; hence we will make in the sequel F = " = 0 w.l.o.g. As in [23], it is assumed that the received signal has been processed by an ideal lowpass filter with cutoff frequency 1+↵…”

“…Characterizing all e h(t) satisfying (4) was first addressed in [23], where the design of e g R (t) was undertaken in order to exploit the spectral coherence of the cyclostationary signal. An alternative description, valid for standard Nyquist pulses h(t) such as the RC pulses considered here, is given next.…”

“…In this model, the signal-free bandwidth available for estimation of the noise variance becomes arbitrarily large as the oversampling factor is increased. In contrast, the NDA estimator in [23] is based on secondorder statistics of the oversampled MF input, but assuming an antialiasing filter removing spectral content outside the signal bandwidth. This seems reasonable, as in practice such content is subject to interference from adjacent channels.…”

Frequency-domain GLR detection of cyclostationary signals in frequency-selective channels

“…In particular, the scheme reported in [10] and method based on [6,11] are considered as benchmarking. The frequencydomain formulation adopted in this work was introduced by the authors in [14] in the SNR estimation problem.…”