Recent advances in software radio, distributed synchronization, and data networks enable the development of coherent software radio arrays. Such systems consist of software defined radios, which are operated with a common distributed time and frequency standard. Networks of coherent software radio systems are a highly flexible class of distributed sensor that can provide wide spatial coverage and high spatial resolution for observation of the Geospace environment. Other examples of passive and active radio arrays for monitoring of the space environment have proven of great utility. For example, Global Positioning System (GPS) monitoring and mapping of total electron content [Mannucci et al., 1998] and distributed HF radar monitoring of ionospheric convection [Greenwald et al., 1995] (i.e., SuperDARN).
 Software radio is an approach to building highly flexible radio systems which can receive and process signals in a manner defined primarily by generally programmable digital elements [Mitola, 1995]. Applications such as spectral monitoring, ionospheric radar, and satellite beacon reception can share a common platform where the primary differences are in the software which implements the necessary signal processing. The computing systems implementing such signal processing can also be shared between multiple instruments.
 Scientific applications of software radio are highly tolerant to latency when compared with communications applications. This difference can be exploited to good advantage. In particular, radio frequency (RF) bandwidths can be manipulated in several key ways. RF data can be buffered locally while lower data rates are processed in realtime to identify interesting periods. Stored voltage level data can also be reprocessed at a later time using improved algorithms and techniques. This processing can take place long after the original voltage level data is captured (i.e., even years, decades, or potentially longer). It is also possible to process data multiple times using iterative or adaptive optimizations. By transporting coherent voltage level representations of radio signals over large distances using digital communication networks, it is possible to synthesize a very flexible instrument. The resulting system is useful for a range of scientific observations using well established techniques as well as enabling new approaches.
 We have implemented the Intercepted Signals for Ionospheric Science (ISIS) array to investigate the utility of a coherent software radio array to study the plasmasphere boundary layer. A key motivation for construction of the ISIS array is to allow the real world evaluation of the effectiveness of a software radio sensor network for observation of the Geospace environment. The Array complements the Millstone Hill UHF Radar System, a 2.5 MW incoherent scatter radar, and can be operated to provide complementary multistatic radar measurements at both VHF and UHF frequencies.
1.1 Coherent Software Radio
 The Coherent Software Radio is the fundamental component of a distributed software radio array. The coherent software radio system consists of an analog interface to the electromagnetic environment and a coherent interface layer, which linearly translates signals to the digital domain with precise global time alignment, high stability in frequency, and minimal addition of noise. This is a key boundary which helps to isolate details of the data acquisition hardware from the signal processing and software. Local computation and buffering memory is available to process signals in real time or store them for later use. A network interface connects the receiver to the wider network and provides for remote control and reconfiguration. These radio systems are combined into a networked array that can operate collectively to make observations.
 The general pattern of a sensor array enabled through distributed coherence [Alexander, 1977] reoccurs in many modern radio and radar instruments. It is by no means unique to Geospace radio science applications and has been widely exploited by other communities (e.g., Radio Astronomy and DoD). Applications of such approaches to astronomical observations are being extensively investigated, and the current generation of distributed radio telescope arrays (e.g., MWA, LOFAR, LWA) exploit this essential architecture in a range of configurations [Bowman et al., 2007; Butcher, 2004]. Active adjuncts to such arrays are also under consideration (e.g., LOIS) [Thidé, 2007].
1.2 Array Coherence
 It is possible to combine the software radio approach with the precision timing possible using signals from the Global Positioning System (GPS) [Lewandowski et al., 1999]. The resulting coherent software radio system can align its data and mode changes very precisely in time and maintain a high stability in frequency even between systems separated by large spatial scales (∼1000s km). This is a similar approach to the use of atomic clocks for very long baseline interferometry (VLBI) derived coherence in radio astronomy observations and has been an important basis for astronomical radio synthesis imaging techniques for many years [Counselman et al., 1977; Clark et al., 1979]. Synchronization of distributed software radio networks must be of sufficient precision to make meaningful comparison of signals between sites in the array (e.g., [Rogers and Moran, 1981]).
1.2.1 GPS Performance and Limitations
 The Global Positioning System (GPS) provides a precise atomic clock broadcast from multiple satellites with known orbits. In the current generation of commercial receivers, the synchronization is limited to pulse per second (PPS) alignments in the range of 20 to 50 ns for high quality units. Lower quality receivers typically achieve on the order of 100 ns. Filtering of the GPS derived PPS signals can improve this to some extent, and stabilization of oscillators on long time scales with Allan variances in the one part in 10E11 or a few parts in 10E12 range can typically be achieved. This translates into networked coherence appropriate for a wide range of ionospheric measurements under most conditions.
 In some cases the performance of GPS as a synchronization source can be degraded. For example, Space Weather events can disrupt the ability of GPS receivers to maintain lock on the satellite signals [Skone and De Jong, 2000; Lambour et al., 2003]. It is also possible for the GPS synchronization method to introduce artifacts which impact the measurements of an array. An example of this is that the filtering involved in oscillator stabilization can introduce unintentional variations in the stabilized oscillators. In some cases, this can be observed in cross-correlated signals as a wandering of frequency alignment between sites over seconds to tens of seconds. The occurrence of such variations are specific to the details of the hardware involved. The effects can be characterized in the laboratory using a modified or dynamic Allan variance [Nunzi et al., 2007; Allan, 1987]. They can introduce artifacts into the measurements made by a coherent radio sensor array.
1.2.2 Self Synchronization Techniques
 In some applications and conditions, it is desirable for an instrument array to achieve coherence using direct reception of signals in the RF environment or to utilize characteristics of observed RF signals to achieve synchronization. Holdover for the case where GPS lock is lost is one example, but some systems might wish to dispense with the cost of GPS synchronization entirely. This is particularly true for very low power networked sensor elements, and a number of possible strategies have been investigated by other authors [Elson and Rōmer, 2003].
 One example of this is the derivation of high precision time synchronization from common pulsar observations [Taylor, 1991; Arzoumanian et al., 1994], but this takes a substantial and flexible aperture. Another approach is to use the narrow and near zero Doppler bandwidth of ground clutter or multi-path observations to correct frequency differences between networked radar systems. Man-made signals will also often contain explicit or hidden synchronization information that can be received and used for self synchronization.
 An alternative approach for a computationally powerful software radio system might be to directly receive GPS signals and then use software based signal processing to create the required system coherence by solving for the synchronization as part of the overall data analysis.
1.3 Wide Area Voltage Level Transport
 A key aspect of a coherent software radio array is the translation of the RF signal coherently into an associated representation as a complex RF digital voltage. Essentially RF bandwidth is translated to data network bandwidth which enables these signals to be transported, analyzed, and compared at a fundamental level. This translation is most general when it is fully coherent in time and frequency and where a detailed characterization of the analog and digital channel between the electromagnetic environment and the conversion point in the system is available. It is also useful to associate metadata information with the voltage level data. For example, such metadata can include information about receiver location, analog path transfer functions, coherence metrics, and specifics of the data acquisition hardware configurations.
 Manipulation and processing of data at the voltage level is fundamental to the multi-role applications of software radio arrays and is also a key to the application of general purpose computing technology for array signal processing. Many of the challenges of this approach (e.g., limited network bandwidth and the required computing power) are being alleviated through technological advances.
1.3.1 Network Transport Considerations
 We first stumbled upon the transport of coherently aligned voltage samples over wide area networks in the early development of the Manastash Ridge Radar system [Sahr and Lind, 1997]. At that time, the fundamental nature of this approach was not clear to us. The available data network was fairly limited in bandwidth (500 kb/s), and only two sites were connected and processed for the resulting passive radar applications. This level of available bandwidth is useful, but can be fairly limiting for wide RF bandwidths sampled at high dynamic ranges. It is currently possible to do better than this in many locations. However, in practice, it can be quite difficult or expensive to achieve high performance networking in remote locations (e.g., dedicated fiber or expensive satellite uplinks). Wireless approaches do offer some deployment advantages for a coherent software radio array, and the potential for self networking arrays is very interesting.
 As networks have become ubiquitous and bandwidths have increased, it has become clear that voltage level RF transport is a key element in enabling highly flexible and distributed software radio sensor networks. In retrospect, a clue to the general importance of wide area voltage transport could have been recognized in VLBI techniques, which evolved from transport of data using tape, then disk, and most recently networked transport of large coherent RF bandwidths with low dynamic range (eVLBI) [Szomoru et al., 2004].
 In a networked coherent software radio array the useable bandwidth for transporting RF signals is in general limited and often asymmetric between sites. This limitation forces selective and intelligent use of the available network bandwidth. Typical bandwidths in the ISIS Array currently range from about 1 to 10 Mb/s with relatively low latencies and good connectivity. This was achieved by preferentially selecting well connected sites.
 In coping with bandwidth limits and asymmetries an obvious approach is to limit array performance in some manner. A simple means to do this is to limit the array duty cycle or operate in a burst mode of operation. Another method is to transport limited bandwidths in order to identify time intervals which would be valuable for intensive processing. Multiple signal distributed compressive sensing [Baron et al.2009] also provides some indication of a useful means of data bandwidth reduction for a distributed software radio array. Applying such techniques in an array wide sense is a future direction for investigation.
1.4 Software Radio Signal Processing
 Software radio offers great flexibility in the signal processing and analysis of radio signals from an instrument array. We have discussed patterns for Software Radar signal processing in previous work [Grydeland et al., 2005], and many of the same approaches and patterns work equally well for networks of software radio systems.
 Signal processing implemented in software using general purpose computers helps to avoid many of the costs associated with the re-implementation of systems due to changes in the underlying hardware technology. It also allows us to take advantage of improvements in computing and networking capacity to increase the capabilities and performance of radio science systems and to share computing capacity between systems.
 This flexibility comes at a price as the initial implementation of the needed software can be complex and limited in performance. The later problem can be mitigated to some extent using specialized processors to accelerate selected operations (e.g., digital filtering) where the implementation cost and difficulty can be kept low. However, for our applications, we have explicitly decided to limit the RF bandwidths that we capture and process to the relatively narrow limits (i.e., kHz to MHz), which can be handled with off the shelf computing systems and networks.