Possible Approaches
Most networking devices today, such as NICs and Wi-Fi components, are built to address a distinct, narrowly defined requirement: communicate over one channel using one protocol as inexpensively as possible. Limited adaptive techniques are employed today (for example, determining the best data rate for communications), although there isn't any switching between protocols/networks. These inflexible devices typically use dedicated hardware.
Some flexibility can be gained by using a DSP and an ASIC. The DSP processes the signal and offloads some processing of protocol algorithms to the ASIC. Other designs use DSPs with closely coupled accelerators. DSPs, however, require lots of power, and this type of radio would need multiple DSPs. All previously mentioned devices have favored a simple hardware architecture of antenna-PHY-MAC that's well established and well understood.
A soft solution, where a reconfigurable processor is employed to handle the necessary protocol, hasn't been viewed as a viable alternative. Such processors don't have the raw processing power or can't reconfigure themselves on-the-fly while handling an ongoing communications protocol. But as ICs have realized the benefits articulated by Moore's Law, the power of reconfigurable processors to execute complex algorithms in real time has seen dramatic increases. As a result, for the first time ever, adaptive radios can be pursued as a viable technology. These adaptive radios will boost spectrum efficiencies by taking better advantage of unused spectrums and determining the optimum spectrum as a function of time, space, and frequency. The result will be better communications for users.
Reconfigurable Communication Core
We've been developing the silicon building blocks for implementing such inexpensive adaptive radios. Current research includes prototype radios that comprise a mesh of heterogeneous processing elements (PEs), with multiple, routable paths between them. Many researchers have chosen homogeneous arrays that permit lots of flexibility, incorporating ALUs and processors at each node.
These arrays typically suffer from ALU and processor requirements of decoding and relaying instructions every cycle. Due to their general-purpose nature and the extreme flexibility allowed, they require significant amounts of memory and cycles to properly compute a given function. The cost in size and power can be several orders of magnitude compared to dedicated logic (Fig. 1). Each tick mark on the MOPS/mW (millions of operations per second per milliwatt) axis represents an order of magnitude. The dedicated logic handles only one protocol, but the coarse-grain and DSP/FPGA logic can support multiple protocols. The difference, of course, is that the DSP/FPGA can handle anything. But with this infinite flexibility comes a severe power penalty, as shown in Figure 1.
The reconfigurable communication core (RCC) is a result of extensively examining many wireless protocols and identifying similar and heavy-computational-burden processing kernels across them. Special reconfigurable accelerators with a much coarser granularity than FPGAs are then designed to execute these functions. Optimally, the functions are reconfigured only once to avoid cycle-by-cycle reconfiguration for each function to be executed, and data is "streamed" into and out of the PEs. Processing is done both spatially and temporally with a significant amount of spatial parallelism.
As much as possible, processing similarity between dedicated hardware processing (complete spatial parallelism) is employed. Time-division multiplexing then is used until the point where frequencies are allowed by the semiconductor process. Thus, processing frequencies aren't being tied to small multiples of the analog-to-digital converter (ADC) rate, as in dedicated hardware. This yields an advantage in area savings that mitigates the increased area due to reconfigurability. Architectures such as RCC hold the promise of approaching the power/size of dedicated hardware while maintaining enough reconfigurability to address most wireless protocols.
For example, the PEs are components like filter-microcoded accelerators, forward-error-correction accelerators plus interface elements, and controller processing elements. The radio can reconfigure itself for various protocols by simply controlling the route that data packets take between the different PEs, as well as reconfiguring the PEs themselves.
The reconfigurable approach conserves power and size compared to the alternative solutions of multiple dedicated cores and other software-defined approaches to radio. In addition, the reconfigurable approach has the eventual advantage of easier programming, along with greater portability and scalability of elements to other platforms and future protocols using existing tools. The scalability is offered via the "building block" mesh connect. This enables connections between "new elements" and their associated communications infrastructure (e.g., routing nodes, etc.), similar to adding new roads when building more houses in a neighborhood (Fig. 2). The new "roads" or interconnect infrastructure also doesn't affect the existing infrastructure. This architecture builds on the fact that for communications PHY applications, the operations are highly pipelinable, and for the most part they require communications only with the nearest neighbor processing element. If interconnect congestion occurs with very large networks of processing elements, it's easy to add an option to add a simple hierarchy to the interconnection architecture.
Continued research is driving further componentization of adaptive-radio elements. The goal is to make them sufficiently effective and inexpensive so that they become the standard solution for constantly-on, high-quality wireless communication in the home, at work, and in between. For this vision to become reality, however, spectrum-allocation policies will need modification as well, so that all possible bandwidth is available for constant communication. The table summarizes the design guidelines for adaptive radios.
Advances in technology are enabling more efficient spectrum usage and influencing reforms in spectrum policy. The results of current research into cognitive, reconfigurable radios will permit ubiquitous wireless communication across multiple protocols, networks, and spectrums.