Collaborative and Responsive Sensors for Low-Cost Spectrum Sensing and Geolocation 2.0 APPROACH TO CLUSTERED SENSOR NETWORKS

Distributed sensor networks can serve as a robust electronic warfare support capability for sensing in contested and congested electromagnetic environments. While the response of an individual node may be unreliable due to interference or poor signal to noise ratio (SNR), the overall sensor network performance is minimally affected. In addition, a distributed sensor network enables geolocation of challenging emitters. Both sensing and geolocation benefit from increased sensor density. Therefore we present a low-cost distributed sensor network for spectrum sensing and emitter geolocation. To meet the low-cost objectives (<$70 per sensor) for this network we relax hardware performance requirements for individual sensor nodes resulting in analog impairments including poor image rejection, local oscillator drift, high phase noise, and limited dynamic range. In this work we show that cross-correlation signal processing can be used to compensate for poor node performance. Many distributed sensor networks rely upon an asymptotic network comprised of a very large number of nodes in order to achieve acceptable network performance in spite of poor node performance. This approximation is not applicable when a limited number of nodes are present, particularly when they are clustered into local groupings. In this work we discuss a non-asymptotic, tactically-relevant sensor network clustered into squads of 5-10 nodes. To ensure overall performance of the network our approach is to connect individual nodes through a backhaul network and operate them in a collaborative and responsive mode. In this context, ”collaborative” implies that nodes are able to communicate locally (in a cluster, e.g., squad) and globally (with a network controller, e.g., platoon), and “responsive” implies nodes are reactive and can change their operating state depending on input from a local node or the global controller to mitigate various analog impairments. We develop a hierarchical signal processing method in which cross-correlation is used at the squad level for improving the performance of a cluster of sensors, and then higher-level algorithms are used at the network level. The method leverages differences in free-space path loss between sensors in a squad to achieve high linearity while maintaining low-noise levels. Performance is quantified through spur-free dynamic range (SFDR). Multiple measurements of a sensor-pair can be averaged to increase SFDR, or multiple sensors can be pair-wise averaged to increase SFDR while maintaining scan rate. We show that cross-correlation is a natural choice for processing clustered networks and that higher network level decisions can optionally adjust the performance of individual responsive nodes by adjusting front-end attenuators. DISTRIBUTION A: Approved for public release; distribution unlimited STO-MP-SET-241 13-4 1 PUBLIC RELEASE PUBLIC RELEASE Collaborative and Responsive Sensors for Low-Cost Spectrum Sensing and Geolocation 2.0 APPROACH TO CLUSTERED SENSOR NETWORKS Low-cost, low-power sensor nodes typically employ swept-tuned receivers with limited instantaneous bandwidth and sample rate to meet those criteria. Therefore, to achieve high scan rates and limit the collection of information to the intra-nodal communication rate, only a small number of time-domain samples are recorded in each tuned band. This, in addition to the analog limitations of such low-cost sensors, motivates the question of this work: “Given a collection (squad) of low-cost, low-power sensor nodes with poor analog linearity, coarse time-synchronization between nodes, and observing approximately similar spectral scenes, how should the squad-level sensor data products be processed?” In the following we compare two popular methods for deriving power spectra based on power-averaged fast Fourier transforms (FFTs) and cross-correlation. We will show that in the context of spectrum sensing of arbitrary emitters cross-correlation is the preferred approach. Further, we will show that in the context of a distributed sensor network where the path-loss between a particular emitter and the various sensors in a squad can vary over many orders of magnitude, this effect can be leveraged to increase linearity. Within the cognitive radio community, the signal to noise ratio wall (SNR wall) theory has found credence and become a source of discussion as to the best way to approach the difficulties imposed by its implications, particularly for the use of spectrum sensing. The SNR wall [2,3] states that for a given uncertainty of the estimate of the noise, there is a minimum SNR (for common sensing applications, -6 dB SNR) at which accurate detection is impossible. It was shown in [4] that cross-correlation between two receivers can overcome the SNR wall with an efficiency related to the degree of noise correlation between the two receiver chains. If two receiver chains measure the same correlated signal but add uncorrelated noise a cross-correlation implementation can be realized. By using the cross-correlation to lower the uncorrelated noise, the uncertainty of the noise is also decreased, effectively lowering the SNR wall. In [5] a highly linear spectrum sensor was proposed wherein attenuation was added to the front-end of one of a pair of on-chip sensors and then cross-correlation was employed to recover noise performance. This was demonstrated in hardware with on-chip cross-correlation receiver pairs [6,7] where the objective is to maintain a very low channel-to-channel correlation coefficient. Here we extend the method proposed in [5] to the case of free space and we demonstrate the approach for more than two sensors. We compare the cross-correlation approach with power-averaged FFTs through simulation and then validate the cross-correlation algorithm through free-space measurements. Here we describe several possible computations and compare their ability to improve overall linearity and noise performance quantified through the spur-free dynamic range (SFDR), defined below. To facilitate this discussion, we define NFFT as the length of an FFT, and M as the number of FFTs averaged together in the following computations: • Coherent Averaging of Complex FFTs: Coherent averaging of M complex NFFT-point FFTs from a single sensor is the most straightforward approach for reducing noise power (-10dB/decade). This is effectively computing a large M*NFFT–point FFT. However, for typical signals of interest (e.g., modulated data or pulsed carriers) this method is not viable because the signals are attenuated along with the noise. • Power Averaging of FFTs: Power averaging of M complex NFFT-point FFTs is similar to coherent averaging but uses the magnitude of the FFT values as opposed to the complex data. This has the benefit of avoiding the destructive self-interference that the coherent averaging sees, but also the inability to reduce the noise. Instead only the variance of the noise floor is reduced which does not improve SFDR with increasing M. • Autocorrelation: Averaging M autocorrelated NFFT-point FFTs is very similar to power averaging of FFTs. Instead of taking the magnitude of the FFT, the power spectrum is found as s*(conj(s))=abs(s)2. PUBLIC RELEASE Collaborative and Responsive Sensors for Low-Cost Spectrum Sensing and Geolocation 13-4 4 STO-MP-SET-241