Stanford-developed Transceiver Operates Full Duplex on a Single Channel

To avoid interference, wireless transceivers can switch between transmit and receive on one frequency (Time Division Duplex (TDD)). Or, they can transmit and receive at the same time on different frequencies (Frequency Division Duplex (FDD)). There’s been a flurry of press reports about a new radio system, developed by Stanford researchers, that can operate full duplex on a single channel; that is, transmitting and receiving at the same time on the same frequency, something not done before.

The reports seem to have been triggered by a February 14 Stanford News Service release. For those wanting to go beyond the headlines, the researchers have a web site and make available a technical paper that was presented, along with a demonstration, at Mobicom 2010 in September. On its face, this system seems to halve the spectrum needed for a two-way system, but it’s not that simple. Furthermore, for reasons you’ll see below, this doesn’t seem to be a mobile solution. The researchers are coming  from the perspective of improving the performance of WLANs, and the paper is more clear when read from that view. Still, no matter what radio system you work with, full-duplex on the same frequency makes you think. I’ve read the paper and have pulled out what I think are the essential points.

The custom has been to not transmit and receive on the same frequency at the same time because it doesn’t work; the receiver is overwhelmed by interference. Interference cancellation techniques that can help, but not enough. To reduce interference to the point where the receiver can detect the desired signal, we get to the novel aspect of this system. The transceiver uses three antennas, two for transmit and one for receive. Power is split between the two transmit antennas. The transmit antennas are placed such that one is one-half wavelength apart from the other, with respect to the receive antenna. The transmitted signals thus arrive at the receive antenna 180 degrees out of phase and cancel, mostly, in a process the researchers call antenna cancellation. After that, RF and baseband interference cancellation reduces remaining interference to the point where the desired signal can be detected.

If that’s too opaque, think of noise-cancelling headphones.

The system, as implemented, has several practical limitations:

  • The two transmit antennas produce a pattern, in the horizontal plane, that varies according to their placement and how they are fed in amplitude and phase. This produces a null (low or no signal) where the receive antenna can be placed, but it also produces undesired nulls where one wants coverage. Adjusting the antennas’ power ratio can fill those nulls to some extent.
  • Null position is sensitive to slight differences in transmit antenna power ratios. If the null moves too much, interference returns.
  • Null position is sensitive to slight differences in antenna placement. At the frequency used for testing (2.48 GHz), if an antenna moves too much — on the order of 1 millimeter — interference can return.
  • The bandwidth is narrow (5 MHz at 2.48 GHz); if the signal is too wide, the outer edges don’t get canceled and there’s interference.
  • The requirement that the transmit antennas be at least one-half wavelength apart means that lower frequencies become awkward to work with. At 2.48 GHz, one-half wavelength is 5 inches. At, say, 700 MHz, it’s 17 inches.

On the plus side, the researchers say this system can alleviate several wireless networking bottlenecks, albeit with reworking of WLAN MAC layers to allow full duplex (which they’re working on).

  • With no time-division, the hidden node problem is reduced since the access point can respond without delay to the first transmitting node. Other nodes hear that response and delay their transmissions, reducing collisions.
  • Full-duplex reduces loss of network throughput cause by congestion and MAC scheduling since congested nodes can send and receive packets at the same time.
  • Delay in multihop networks is reduced because a node can start forwarding a packet as it receives it, instead of using typical store-and-forward techniques.

They also point to a potential application in cognitive radio; a secondary user, while transmitting, could monitor for the primary user. In addition, the ability to have a control channel in-band and in real-time raises the prospect of improving the performance of some systems.

The full-duplex prototype, made with off-the-shelf parts and incorporating the IEEE 802.15.4 modulation/demodulation scheme, achieves performance within 8% of an ideal system. Some of this shortfall is caused by granularity of the test setup, such as using attenuators with larger-than-desired steps. The researchers are considering applying the technology to IEEE 802.11 radios; that’s a challenge because both power and bandwidth are larger (more interference to be suppressed).

The system seems to perform best with single propagation paths. In the presence of multipath, I’d expect a reduction in performance due to fading; the researchers report multipath was not a “dominant component” in their tests, which were done indoors with pretty good results. I’d like to see simulated or measured performance under a few different multipath conditions.

We handle multipath on WLANs well today through the use of multiple-input and multiple-output (MIMO) antennas. With sufficient multipath, and a sufficient number of antennas, spectrum capacity can be doubled, or more. But MIMO in the WLAN context doesn’t permit full duplex, and thus doesn’t permit the networking fixes that this system does, and those improvements are expected to be a significant source of gain. Furthermore, MIMO is designed to take advantage of multipath; its gain isn’t that great where multipath is low, such as on some outdoor links.

The tradeoffs of this system when comparing it to others are gains from reduced spectrum requirements, losses from lack of MIMO, and gains from relieving the hidden node problem, reducing network congestion, and reducing end-to-end network delay. The researchers suggest the biggest benefits are to come from reducing network bottlenecks; they downplay physical layer gains.

We’ll follow this. Reading about it reminds me of sitting in a wireless standards meeting in the mid 1990’s and hearing about Turbo Codes, a coding scheme that doubles data rates with no increase in transmitted power, and that is in widespread use today. That concept also came out of the blue. Many were skeptical, but it worked. As with this radio system, Turbo Codes were made from existing elements put together a different way. Sometimes that’s all it takes.

(Disclosure: According to the paper’s acknowledgments, this research is supported in part through a gift from DOCOMO Capital, a subsidiary of NTT DOCOMO, which is a client.)