Marc Willerton - Imperial College London, Communications and Signal Processing Group
David Yates - Imperial College London, Circuits and Systems Group
The Need for a Reconfigurable Antenna Array Testbed
Antenna arrays exploit spatial diversity to enhance performance in signal detection, parameter estimation, and reception for further capabilities such as beamforming, direction finding, target localization, tracking, and radar. These techniques can be used in a variety of applications, including military, industry, commerce, and health. Researchers recently developed novel array-processing techniques that, in principle, offer significantly superior performance. However, there often is a significant lack of opportunity for communications research groups to prototype these algorithms using real hardware due to the high cost and application-specific nature of arrayed RF front ends. The result is often a missed opportunity to use promising new algorithms in real-world systems.
To tackle this problem, the Communications and Signal Processing Group at Imperial College London teamed up with the Circuits and Systems Group at Imperial College London to develop a low-cost, reconfigurable antenna array testbed suitable for the rapid prototyping of a wide range of new array-processing architectures and algorithms. The project forms part of the University Defense Research Centre in Signal Processing workfunded by the UK Ministry of Defense and the Engineering and Physical Sciences Research Council.
To test the wide range of algorithms developed for a large variety of applications, any useful testbed must be highly reconfigurable in terms of frequency band, bandwidth, and number of array elements, with each channel possessing both transmit and receive capabilities. In general, the portability of the array is crucial for testing in a wide variety of environments, and for some applications, such as wireless sensor networks and bistatic multiple input, multiple output (MIMO) techniques, arrayed transceiver channels should be widely distributed in space. Successful array operation requires frequency, phase, and timestamp synchronization. In addition, large bandwidth is often beneficial for improved performance. The processing of raw data in real time aids the rapid prototyping of novel algorithms so that we can easily evaluate and improve performance. We often prototype algorithms using a PC interface with .m file syntax compatible with The MathWorks Inc. MATLAB® software and the LabVIEW MathScript RT Module.
High-performance phase-aligned RF systems are typically priced beyond the budgets of academic researchers developing these algorithms, and many of these systems don’t offer the degree of reconfiguration desired. The research group previously developed a custom RF receiver based on Mini-Circuits components that was used with an NI PXI-5105 digitizer as an array testbed. However, the RF front end suffered from poor noise performance, had receive-only capability, and was not reconfigurable in terms of frequency band, number of receivers, and location of array receiver channels. In contrast, the NI USRP platform combined with LabVIEW software provides a cost-effective approach, with the potential to meet the above requirements. As a phase-coherent software defined radio, our solution needed phase compensation to achieve phase alignment. Therefore, much of this research project involved developing the synchronization and calibration techniques necessary for individual NI USRP boards to successfully operate as a phase-aligned array.
Array Testbed Setup
Each NI USRP board is a single-element transceiver that can cover a frequency range from 50 MHz to 5.9 GHz, depending on the model chosen. The array testbed setup at Imperial College London consists of up to 12 NI USRP radios covering the 2.4 GHz industrial, scientific, and medical band (Figure 1). Each board provides one array channel with the antenna connected to the RX1 port. Boards are interfaced with the PC via an Ethernet switch, and each board features a common 10 MHz clock reference signal and a 1 pulse-per-second (PPS) signal. This means we can tune the local oscillators on each board to the same frequency and helps with timestamp synchronization. However, this alone is not enough to form a phase-aligned array system. In addition, the phase ambiguities created when the local oscillator is derived on each radio from the 10 MHz reference clock must be calibrated out. Each time a retune command is sent, the phase between channels changes because of this ambiguity. For phased-array applications where the phase difference is key, this problem must be compensated for each time a new set of data is collected. Phase synchronization is performed by applying a common 2.401 GHz tone to the RX2 port of each board in the array via a splitter. This signal is separated from the over-the-air signal using a digital filter, and it is used to estimate phase ambiguities in the boards. These ambiguities are then applied as phase correction to the over-the-air signals to provide a synchronized array system.
Testbed Use Cases
A number of so-called “superresolution direction-finding and localization algorithms” have been implemented on the array NI platform-based testbed. These algorithms can asymptotically achieve zero errors under ideal statistics. In addition, they offer enhanced performance, compared to using beamforming for the purposes of localization, because they don’t suffer from the effects of sidelobes. This means we can greatly reduce the number of elements in the array for a given directional resolution. However, such algorithms suffer from large degradation in performance in the presence of array uncertainties; hence, array calibration becomes an important issue to address. Common array uncertainties include gain and phase uncertainties associated with the array elements, and uncertainties in the array geometry. We devised pilot-based array calibration algorithms to attempt to compensate for these effects. Here, the responses of pilot sources transmitting from known locations are used to infer the array uncertainties by solving a set of linear equations.
Case A: The MUltiple SIgnal Classification (MUSIC) Algorithm
The MUSIC algorithm is a well-known, subspace-based, direction-finding algorithm . Consider a uniform linear array of N=4 sensors. A single source operating at 2.43 GHz is located at 32 deg in the far field of the array whose location must be estimated using the MUSIC algorithm. Then, L=100 snapshots of data are collected from the array within an anechoic chamber under a signal-to-noise ratio (SNR) of X dB. A pilot source is located in the far field of the array at 105 deg to perform gain and phase calibration. Following this, direction finding is performed, giving the result in Figure 2. After the calibration process, the performance of the direction-finding algorithm is within 2 deg of the true source direction.
Case B: The Large Aperture-Array Localization Algorithm
Direction-finding algorithms such as MUSIC use small aperture arrays to provide a direction estimate of a transmitting source. Hence, multiple array systems are required to localize the source. In contrast, we designed a large aperture-array for localization using a different algorithmic approach using a single array with elements spaced over a large distance. This large aperture requirement helps us take advantage of the flexibility of the array testbed based on the NI USRP platform.
Consider a wide aperture array of N=4 sensors with the geometry shown in Figure 3. A source operating at 2.43 GHz is placed between the array elements whose location is to be determined. Again, L=100 snapshots of data are collected from the array within an anechoic chamber under an SNR of 35.14 dB. A pilot source is placed in the field to estimate the gain and phase uncertainties associated with the array. The result after the calibration process is shown in Figure 4. The system is able to obtain a low localization error of approximately 6 cm. More practical results associated with this work can be found in the paper “Experimental Characterization of a Large Aperture Array Localization Technique Using an SDR Testbench,” published at the Wireless Innovation Forum Conference on Communications Technologies and Software Defined Radio (SDR’11-WInnComm) .
The LabVIEW and NI USRP platform provides a low-cost, reconfigurable antenna array testbed suitable for the rapid prototyping of algorithms developed at Imperial College London. The results obtained from the testbed so far are promising, and we look forward to expanding our research with more sophisticated array calibration procedures developed by Imperial College London. We are hopeful that the performance of this system will further improve as we compensate for inconsistencies between antennas and antenna placement. This and other array-processing algorithms, including those using the array to transmit, will be investigated in the future.
MATLAB® is a registered trademark of The MathWorks, Inc.
- 1] Schmidt, R., “Multiple Emitter Location and Signal Parameter Estimation,” IEEE Transactions on Antennas and Propagation, vol. 34, pp. 276-280, Mar. 1986.
- 2] Manikas, A., Y.I. Kamil, and M. Willerton, “Source Localization using Large Aperture Sparse Arrays,” IEEE Transactions on Signal Processing, 2012, to appear.
- 3] Willerton, M., D. Yates, V. Goverdovsky, and C. Papavassiliou, “Experimental Characterization of a Large Aperture Array Localization Technique Using an SDR Testbench,” Wireless Innovation Forum Conference on Communications Technologies and Software Defined Radio (SDR’11-WInnComm), 2011.
Circuits and Systems Group
Imperial College London
Imperial College London, Communications and Signal Processing Group