Observaciones y resultados

This work was supported by the Armaments Corporation of South Africa (Armscor) under contract no. KT521896. The authors would like to thank the anonymous reviewers for their valuable inputs.

PUBLICATION 3

This chapter contains the authors’ version of a postprint of a paper submitted to and accepted for publication in IET Communications and is subject to Institution of Engi- neering and Technology Copyright. The copy of record is available at IET Digital Library at www.ietdl.org. The bibliographic details of the paper are given below and in the Ref- erence list as [17].

Title Blind sequence-length estimation of low-SNR cyclostationary sequences

Authors J.D. Vlok and J.C. Olivier

Journal IET Communications

Publication date 12 June 2014

Volume 8 Issue 9 Pages 1578–1588 doi 10.1049/iet-com.2013.0616 Print ISSN 1751-8628 Online ISSN 1751-8636

The submission timeline is given below.

Original manuscript submitted 15 July 2013

Decisioned: Revision requested 25 November 2013

Revised manuscript submitted 23 December 2013

Decisioned: Accepted 17 February 2014

Blind sequence-length estimation of low-SNR

cyclostationary sequences

J.D. Vlok1 _{andJ.C. Olivier}2

1_{Defence, Peace, Safety & Security (DPSS), Council for Scientific and Industrial Research (CSIR), Pre-}

toria 0001, South Africa

2_{School of Engineering, University of Tasmania, Hobart 7005, Australia}

E-mail: [email protected]

Abstract:Several existing DSSS detection and estimation algorithms assume prior knowl- edge of the symbol period or sequence length, although very few sequence-length estima- tion techniques are available in the literature. This paper presents two techniques to estimate the sequence length of a baseband DSSS signal affected by AWGN. The first technique is based on a known autocorrelation technique which is used as reference, and the second technique is based on PCA. Theoretical analysis and computer simulation show that the second technique can correctly estimate the sequence length at a lower SNR than the first technique. The techniques presented in this paper can estimate the sequence length blindly which can then be fed to semi-blind detection and estimation algorithms.

5.1

Introduction

In non-cooperative reception, blind detection and estimation techniques are required as the parameters used by the communication transmitter are in general not known by the receiver. The normal approach used in cooperative communication systems, such as optimal correlation or matched filtering techniques [26], is therefore not applicable in non-cooperative communication receiver systems. Typical applications of such systems are to be found in spectrum surveillance and electronic interception, but blind estimation techniques are important in their own right also [15], and hence this paper considers some aspects of this problem in depth.

Signal detection and parameter estimation are usually performed separately and indepen- dently in communication problems [129]. Detection is performed to determine whether the signal of interest is present or absent given observed data that is corrupted by noise. When it has been determined that the signal of interest is present, estimation is performed to determine the signal parameter values. This paper is concerned with blind estimation (assuming the signal is present) of the period of cyclostationary sequences used in DSSS

The spreading codes used in DSSS are cyclostationary, since the mean and autocorrelation of the transmitted signal are periodic with the same period [6]. This periodicity distin- guishes the signal from noise and can be exploited to perform detection and estimation especially when the signals are weak. This is in fact the case for received DSSS, especially for a non-cooperative receiver which does not know the spreading code and cannot there- fore take advantage of the processing gain. Furthermore, modern communication systems use feedback to limit transmit power levels. The resultant effect is that intercepted DSSS signals have very low SNR levels and sophisticated algorithms are required to perform detection and estimation reliably.

We wish to emphasise that blind estimation of the sequence or code length (or symbol period) of hidden DSSS transmissions is essential since semi-blind techniques often assume knowledge of the sequence length which is generally not known a priory. Such semi-blind techniques include sequence estimation techniques [77, 86, 93] and detection algorithms [16].

A few techniques that may be used to estimate the sequence length (or related parameters from which the sequence length can be determined) of DSSS transmissions have been suggested in the literature. These include cyclic-feature analysis to determine the chip rate through spectral-line regeneration [4], the related Fourier analysis of cyclic correlation to estimate the bit rate [86], and higher-order statistical analysis where unique bispectrum and triple correlation patterns reveal characteristics of m-sequences that may be used to determine the sequence length and generator polynomials [111]. These techniques however require relatively high SNRs to perform parameter estimation. Another technique based on autocorrelation suggested in [54, 55] has the potential to estimate the sequence length at lower SNR values.

In this paper, we propose two new methods to estimate the sequence length of an inter- cepted DSSS signal at low SNR, and we present some novel results based on these new methods. The first method is based on the autocorrelation technique [54, 55] mentioned above, and the second method on a PCA detection technique [16]. The two techniques are compared in an AWGN environment in terms of the probability of correct estimation of the sequence length.

The paper is organised as follows. Section 5.2 presents the communication and intercept systems, and defines the SNR regime to which the estimation is applied. The two new methods are introduced in Sections 5.3 and 5.4. Section 5.5 presents numerical results, and Section 5.6 concludes the paper. Possible future research areas are identified in Section 5.7. A Barker code (length N = 11) and an m-sequence (N = 63) are considered as test cases throughout the paper.