2. Picasso: Flexible RF and Spectrum Slicing [HotNets'11][SIGCOMM'12]

Steven Hong, Jeffrey Mehlman, Sachin Katti

Picasso Presentation at SIGCOMM 2012

Picasso Demonstration (Click Here for Demo Description) at SIGCOMM 2012

The networks around us are becoming increasingly dense. For instance, in the past a WiFi Access Point may have been expected to maintain only a single connection, but today it must be capable of supporting several. Similarly, the networks around us are becoming increasingly rich as well. Our personal gadgets increasingly play host to a rich and diverse set of concurrent wireless applications. A single device may wish to simultaneously access the Internet via WiFi, stream video and audio directly via WiFi-Direct from peer devices, and even triple up as a game controller to a gaming console via WiFi-Direct.

Meanwhile, a conflicting course is trending in mobile device design: the real-estate allocated for antennas and other RF components is becoming more and more limited. Equipment manufacturers are increasingly favoring slimmer profiles and larger battery sizes, leaving precious little room to support the radio on each device. However, portable consumer devices such as smartphones must accommodate a growing list of separate protocols (For instance in the ISM band alone - WiFi, WiFi-Direct, GPS, NFC, and Bluetooth need to be supported). Current practice is to use a separate radio and antenna for each protocol, shown in Fig. 2, but as the number of radios increases it becomes difficult to find enough space to separately place all the antennas these radios would need (e.g., the iPhone 4 "antenna-gate" was caused by antennas placed too closely). Is it possible to design a system capable of supporting multiple concurrent protocols whilst using only a single radio and antenna?

The key obstacle is that current radios cannot simultaneously transmit and receive on different arbitrary spectrum fragments with a single, shared Radio Frequency (RF) front end and antenna. The reason for this is that the transmitted signal causes high-powered self-interference, which saturates the RX chain and Analog-to-Digital Converter (ADC), consequently nulling the received signal. While the standard solution is to utilize static, analog RF filters to eliminate the self-interference, such an option is infeasible because spectrum fragmentation is dynamic - available spectrum in the ISM band varies in space and time, depending on the presence of other wireless networks. Consequently, if a radio wants to leverage all the available spectrum and be able to simultaneously transmit and receive on different fragments, the shared analog front end would need programmable analog filters that can be dynamically configured to let only the received signals through and filter out the self-interference. Analog filters however are typically statically configured and programmable analog filters that can be changed dynamically are expensive, lower performing, and impractical to deploy in current radios.

At SIGCOMM'12, we will be presenting the design and implementation of Picasso, a novel full duplex circuit design that sufficiently cancels (instead of filters) the self-interference in analog and prevents RX front end and ADC saturation, enabling the radio to cleanly recover the received signal. Our key contribution here is a circuit design that (1) isolates TX and RX signals at a single antenna by incorporating a circulator [24], and (2) exploits the fact that the self-interference signal travels through the fixed, known circulator channel to design a passive self-interference cancellation circuit. This allows a radio to simultaneously transmit and receive on arbitrary spectrum fragments even while using a single RF front end and antenna. Our design improves on all prior related work on full duplex wireless since they require at least two antennas (one for TX, one for RX), and these need to be separated by 15 to 20cm, which is untenable for small personal gadgets such as smartphones or tablets.

Picasso leverages this full duplex capability to build an abstraction that allows one to flexibly slice a single radio and antenna into separate independent slices operating on different spectrum fragments. Each slice is associated with a specific spectrum fragment in the ISM band (whose width/position can be programmatically specified). The key property is that the operation of each slice is decoupled from the other slices, i.e., the slice is free to run whatever narrowband PHY and MAC protocols it chooses, and the protocol behavior is not impacted by any other slice that may be present on the shared radio and antenna. Thus, in the above scenarios, the AP would have two slices corresponding to the two spectrum fragments and run two independent WiFi OFDM/CSMA protocols on the two slices in parallel. Similarly, a radio could be shared amongst multiple protocols (e.g. WiFi, Bluetooth, and NFC) by assigning independent slices to the appropriate spectrum fragment and running the corresponding protocol.
Further, to ensure that each slice can use existing, well-engineered narrowband PHY and MAC protocols on each slice, Picasso includes a reconfigurable filter engine that transparently takes signals spread over different spectrum fragments, and efficiently filters and resamples them so that the higher layers just see a simple sample stream consisting of narrowband digital samples. The higher layers are then free to process these samples with any narrowband PHY technique they choose, and schedule access to the slice with a MAC protocol of their choice. The slice, for all intents and purposes, appears as their own piece of spectrum centered at zero, operating on their own radio. Picasso thus completely abstracts out the complexity of spectrum fragmentation.

Prototype: We design and implement a prototype of Picasso on Xilinx Virtex-5 FPGA-based software radios. Our implementation consists of both the radio design that provides the slicing abstraction, as well as a WiFi-like contiguous OFDM PHY and CSMA MAC to operate on top of the slices. We show that Picasso's implementation of the slicing abstraction provides strong decoupled operation, i.e. there's no SNR loss because of either programmable filtering or simultaneous TX/RX. In other words, a Picasso radio achieves the same throughput as one would have achieved by using several independent radios statically configured to operate on individual slices.
Application in Dense, Rich Networks: In the paper, we demonstrate how Picasso helps tackle coexistence by using the example deployment shown in Fig. 1. The deployment paints a typical home network: a WiFi AP aims to provide Internet access to a laptop, a tablet, an Xbox and a smartphone. Concurrently, there is a multiplayer game going on, and the Xbox wants to stream high definition individual gameplay video to the smartphone and tablet, and common gameplay to a TV. The smartphone and tablet also double up as wireless gameplay controllers. Further, with the emergence of P2P technologies such as WiFi-Direct (a new WiFi standard for directly connecting two devices without incurring the overhead of an AP), except for the WiFi network, all the other connections will likely be connected via independent WiFi-Direct links. We show that compared to current state-of-the-art coexistence mechanisms, Picasso can provide higher aggregate bulk throughput, improved video quality, and reduced latency simultaneously.

Application in WiFi Access Points with Legacy Clients: We have also implemented Picasso in a WiFi Access Point, servicing legacy clients which are limited to the standard single contiguous channel. In this scenario, Picasso can be used for spectrum aggregation in fragmented ISM bands as shown in as shown in Fig. 5. A WiFi AP can run independent OFDM PHY and CSMA MAC protocols on multiple WiFi channels to simultaneously serve multiple legacy WiFi clients assigned to different channels and achieve significantly higher throughput than a legacy AP that is restricted to use only one channel at a time. Similarly, a WiFi client radio with such a capability can use it to simultaneously connect to multiple WiFi APs on different channels and obtain a much higher aggregate throughput than current radios that can transmit or receive on only one channel at a time.

Compared with a legacy WiFI Access Point, we found that Picasso provides at least a 3X aggregate throughput increase and approximately a 2X gain in median per user throughput. The reason is of course that a Picasso-AP can take advantage of all the available spectrum fragments, while the legacy AP is stuck using the largest fragment available. Further, as expected the throughput drops precipitously for the legacy AP as the number of clients increases. The reason is that the 802.11 CSMA mechanism doesn't scale well with increasing number of clients, the overheads of backoffs and collisions result in even the available largest spectrum fragment being used inefficiently. On the other hand, with a Picasso-AP, since the clients are divided into separate groups corresponding to the different slices, the effect of CSMA inefficiencies is mitigated.

Steven Hong, Jeff Mehlman, Sachin Katti, "Picasso: Full Duplex Signal Shaping", In ACM HotNets 2011, Cambridge, Massachusetts, USA

Steven Hong, Jeff Mehlman, Sachin Katti, "Picasso: Flexible RF and Spectrum Slicing", In ACM SIGCOMM 2012, Helsinki, Finland.