In this post we present ‘Design and Implementation of Wideband Spectrum Sensing on SDR Platform With Receiver Calibration’. The content of this blog is taken directly from our paper which you can download here.

Abstract

This report examines detection algorithms and their implementations for wideband CR network using Nutaq SDR platform. The sensing design and implementation approach employs the algorithms of

[1] which applies ratio based test statistics for detecting the edges of all sub-bands, energy comparison approach to reliably detect the reference white sub-band and generalized energy detection to detect each of the sub-bands other than the reference sub-band. In particular, this report compares the performances of these algorithms with and without performing calibration (which will be clear in the sequel) of the Nutaq SDR platform. Through extensive experiment we have found that performing receiver calibration is helpful to achieve the theoretical performances claimed by [1] reliably. Moreover, we also verify that the considered detection algorithms are indeed robust against noise variance uncertainty and do not suffer from SNR wall. And the performances of these algorithms will not be affected by carrier frequency offset and moderate interference signal, and achieve promising result in the real world scenario.

Introduction

With the popularization of high bandwidth wireless devices in our daily lives, the radio frequency spectrum has become a very precious resource. It is not unusual to find a geographical spot in which some frequency bands (i.e., WiFi bands) are so crowded. On the other hand, there are frequency segments that are inefficiently exploited and only used by regulated conventional networks (for example a public TV service). The concept of Cognitive Radio (CR) has been proposed as a solution to enable the reuse of the underutilized frequency bands in an opportunistic manner. One of the fundamental requirements of a CR device is its capability to discern the nature of the surrounding radio environment to exploit the available spectrum opportunities [2]. This is performed by the spectrum sensing function of the CR network.

Recently wideband spectrum sensing has received a lot of attention where the considered band has more than one sub-bands. Thus, in a wide band spectrum sensing, it is required to determine the number and bandwidth of sub-bands, and examine each of the sub-bands to verify whether it is occupied by the primary user (PU) or not. In [3], [4], the edge detection approach is applied to identify the edges of each sub-band. Once the edges of each sub-band is determined, each of the sub-bands is examined independently using the well know energy detector. The approach of these papers, however, does not provide any analytical approach to examine the performance of their edge detector as a function of signal to noise ratio (SNR). Furthermore, these papers apply the conventional energy detection algorithm directly by assuming that the minimum average power of all sub-bands is equal to the noise variance.

Recently a unified sensing and optimization framework is proposed in [1] (by NECPHY-Lab, INRS) for wideband CR networks with noise variance uncertainty. The paper proposes a simple ratio based edge detection algorithm to detect the edges of each sub-band. And for the given edge, new generalized energy detection algorithm is proposed. The performance of the edge detection and generalized energy detection algorithms of [1] have been quantified analytically as a function of SNR and are shown to be robust against noise variance uncertainty. Furthermore, this paper exploits the fact that ”when the noise variance is estimated from a finite bandwidth, the theoretical results of the conventional energy detection algorithm can not be applied directly”. The theoretical results of [1] are demonstrated by computer simulations. The work of this paper, however, does not provide experimental results to validate the theory.

The current report studies the algorithms of [1] and provides extensive experimental results using a commercial Software Defined Radio (SDR) platforms. In particular, we exploit the usefulness of performing receiver calibration prior to the signal detection phases. And we validate that performing simple receiver calibration helps to improve the detection performances of the system. Furthermore, the experimental results show that the analytical expressions shown in [1] can be obtained in a practical scenario, accurately meeting the theory.

This report is organized as follows. Section II discusses the system model and problem statement. The considered ratio based edge detection, reference white sub-band detection and generalized energy detection algorithms are discussed in Section III. In Section IV experimental results are presented for several practically relevant settings. Finally conclusions are drawn in Section V.

System Model And Problem Statement

Consider a wide band CR network that operates within a given bandwidth of B Hz where different sub-bands have different power spectrum density (PSD).

Fig. 1. The power spectral density of a wide-band signal comprising different sub-bands (SBs): SB2 and SB4 are white spaces.

Fig. 1 illustrates a typical utilization pattern of the spectrum where the number of sub-bands is 5. We assume that a cognitive device attempts to identify the available spectrum holes in order to perform transmission on these spectrum holes1. It is assumed that the considered wide-band contains one or more white sub-bands. This assumption is reasonable since according to the Federal Communications Commission (FCC) report, spectrum utilization on most available frequency bands is quite low [5]. And we consider that the CR network performs sensing and transmission repeatedly over equal frame intervals as shown in Fig. 2.

Fig. 2. The frame structure of a cognitive radio network.

As we can see from this figure, each frame of duration Tf has both sensing time (T) and data transmission time (Tf – T). The sensing time is required to reliably detect the white spaces of the considered wideband whereas, the data transmission time is required to execute data transmission on the white spaces. In practice Tf could be set as the channel evacuation time of the CR network. For example, Tf can be set to 2 second according to the 802.22 standard. We assume that the edges of each sub-band remain unchanged for L consecutive frames. This assumption is reasonable as boundaries of the sub-bands would not change very quickly. Therefore, L can be quite large in practice. However, in these frames, the PSD of each sub-band may change from one frame to another (i.e., a sub-band may contain noise only or signal plus noise in two consecutive frames). Under these practically valid assumptions, the wideband sensing design addresses the following requirements:

Req 1: Detecting the edges of each sub-band from the received samples of the current frame and previous L-1 frames (i.e., This requirement will employ the samples of duration LT seconds).

Req 2: Determining the reference white sub-band from the received samples of each frame which depends on T (i.e., the sub-band which contains noise only). This step is required to estimate the noise variance from the reference white sub-band.

Req 3: Detecting each of the remaining sub-bands (i.e., sub-bands other than the reference white sub-band) using the noise variance obtained from Req 2 and the received samples of each frame which depends on T. In particular, this report examines the effect of L and T on the performance of the edge detection (i.e., Req 1), and the effect of T on the performances of Req 2 and Req 3 using the Nutaq SDR platform (see fig. 2).

Note that in Fig. 2, the target frame (i.e., L) utilizes the samples of the previous L – 1 frames and its own frame. As will be clear later, it is still possible to perform edge detection just from the current frame. However, such approach will not ensure reliable results. Due to this fact we utilize the samples of the previous L frames for our edge detector. By doing so, we are able to identify the edges of each sub-band without any error. From this figure, one can understand that any frame i will utilize the received samples of the frames i – (L – 1) to i.
The content of this blog is taken directly from our new whitepaper which you can download here.