Full-Duplex Wireless Remains A Promise And A Challenge

Wireless bandwidth capacity could double if we could use full-duplex communication. Unfortunately, that’s not as easy as it sounds. People are working hard on it, but we’re mostly not there yet.

Full duplex seems easy. In fact, so easy that it was possible with the first basic telephones. There was one loop between each phone and the central office, and that one loop carried conversations in both directions. Why didn’t we hear both conversations in our ear? Because we simply subtracted out what we were saying on the transmit side — we had a perfect signal to subtract — and then we were left with the received conversation. (We left a bit of the transmitted signal in our ear because it sounded better than dead air.)

So if we could use a single channel for both transmit and receive way back in the early 1900s, what’s keeping us from doing it with our modern phones? And not just phones — WiFi, Bluetooth, ZigBee, and anything that sends and receives signals over the air.

As Gent Paparisto, solutions architect for the AWR Group at Cadence Design Systems, described it, wireless technology separates transmit and receive signals by one of two ways. With frequency-division duplexing (FDD), two separate bands are allocated, one for transmit and one for receive. With time-division duplexing (TDD), by contrast, a single channel alternates in time who gets the channel, the transmitter or the receiver. Whether we use FDD or TDD, we literally divide the bandwidth between the transmitted and received signals. With full duplex, using the same frequency band for both transmit and receive full-time, we could double our available bandwidth.

Paparisto noted there are certain systems that want to avoid the separation. While pulsed Doppler radar sends a pulse and awaits a return, new radar for use in applications like automotive are constantly sending, so they can’t make time for a return signal on the same channel. TDD would not work there.

Yet another benefit would be a reduction in the number of filters in a cell phone. “Right now, there are a lot of filters in phones. At a given frequency range, you’d need only one filter [for transmit and receive] instead of two [one for each],” Paparisto said.

Transmitting loudly while receiving a quiet signal
Shawn Carpenter, Ansys senior product manager, noted that the big difference between wireless and the original phones is literally the wire. A wire is a clean, controlled, stable, well-characterized channel. The signal can travel the distance with minimal attenuation. “With wireless signals, the strength of the signal attenuates proportionally to the square of the distance it travels,” said Carpenter. “You’re transmitting a strong signal while trying to receive a signal that might be 10 orders of magnitude weaker than what you’re transmitting.”

Paparisto added that, with the plain-old telephone system, everything was analog. “The [propagation] loss is much less in wire or fiber. With wireless, if you transmit at 30 dBm, you might have a received signal at -90 dBm. That’s a 120-dB difference.”

Echo, or multipath issues, are other reasons why this isn’t like an old phone. The transmitted signal may travel a circuitous path to reach its intended receiver, perhaps bouncing off of some buildings or walls to get around other obstacles. But those reflections also come back to the transmitter. So it’s not enough to subtract out the transmitted signal because it’s going to be coming back as a reflection, and potentially many times. Longer reflections may be attenuated, but indoor obstacles like metal furniture may cause a quick, strong reflection right near the transmitter.

There are several approaches being worked on to solve this challenge. The first is the most obvious, namely isolation between the transmitting and receiving antennas. “In some cases, transmit and receive might share an antenna, being separated right afterwards,” said Paparisto. “But in many cases, the physical antennas are different.”

The next technique is straightforward — RF or analog cancellation. “You get an idea of the transmitted signal and then play with the phase and amplitude. There are a lot of techniques for doing this,” he said.

But the benefit also is limited. “This gets you a third to a half of the dynamic range needed,”Carpenter noted. “It works for narrowband, but wider bandwidth can be harder to cancel.”

The next technique is digital cancellation. This involves converting to the digital domain and using DSP algorithms to calculate the cancellation. “You have to do this along with RF cancellation,” Carpenter said.

Both together work better than either alone. “You get a lot more power to cancel out the signal” in the digital domain, Paparisto said. “But it takes lots of computation.”

This combined approach is leveraged by Kumu Networks with its self-cancellation technology. The company supplements digital techniques with analog filtering. “Digital is usually the right way to implement thinks, if you can do it,” said Joel Brand, vice president of product management at Kumu. “There are typically more people and organizations who can develop digital solutions. Digital solutions can be better miniaturized. Cancellation, in most cases, cannot be done completely in the digital domain. We are not trying to complete with digital solutions. We are trying to cancel just enough in the analog domain, such that digital solutions (from us or someone else) could kick in.”

Fig. 1: The yellow line is the original signal; the blue shows analog cancellation; and the green shows digital cancellation for a total of about 85 dB. Source: Kumu Networks

Kumu does some of the analog cancellation at the carrier frequency and some of it at an intermediate frequency. “We have to cancel multipath reflections that might arrive 1.4 μs later,” Brand said. “We use a FIR filter to handle multiple reflections during that period.”

The company claims 85 to 90 dB of cancellation, with about 50 dB of cancellation in the analog domain and another 50 dB in the digital domain.

Fig. 2: Kumu Networks splits the energy between carrier-frequency filtering and intermediate-frequency filtering. Source: Kumu Networks

But even this may not be sufficient for full duplex. “To operate in full-duplex mode, one needs to cancel the output power of the radio down, almost to the noise floor (NF). If we take, for example, a 1-W radio (+30 dBm) with -100-dBm NF, you would need 130 dB of cancellation. This is just an example. You can adjust these numbers to any power level and bandwidth,” added Brand.

For very specific applications, Kumu has achieved full duplex. “To augment our [roughly] 100 dB of cancellation, we further isolate the transmitter and receiver by using two separate antennas with 30-dB isolation between them,” he said. “This is how we operate our relays and IABs [integrated access and backhaul] in full-duplex mode.” Such use of full duplex was originally requested by operators for easy-to-install boxes used to expand cellular capacity temporarily, such as a special event or an emergency. Now, the company is starting to implement full duplex more broadly — not between the phone and the cell tower, but in the relays and IABs.

There are still applications where cancellation short of full duplex is useful, he noted, particularly when separate radios are located nearby and operate on nearby frequencies. “This is a common problem on airplanes where satellite communication and radars are operating in nearby frequencies,” he said. So the more common application of self-cancellation is to avoid interference between different protocols in the same bands, like WiFi and Bluetooth.

Leveraging MIMO
To get closer to full duplex, Ansys’s Carpenter identified a third cancellation technique to be added to the RF and digital cancellation — MIMO (multiple input/multiple output antennas) or antenna cancellation. This is of particular interest in the 5G realm. The idea is most easily described with two antennas. If both transmit the same signal, then those two signals will interfere in space. The location of nulls or nodes in the combined signals will depend on the phase relationship between the two antennas.

Ideally, you want the local receiver to be in a place where the transmitted signals cancel. When you have a static physical arrangement, which is typical with WiFi, you can get more cancellation by placing the local receiver in one of these nulls. This also is also referred to as “spatial nulling,” but it may not be easy. And things get more difficult for mobile technologies like 5G.

In order to communicate with some far-end subscriber, the 5G base-station antenna will measure the signal it receives from the far-end over an uplink channel, which is hopefully relatively close to the transmit frequency. Based on that signal, it will calculate how best to direct its transmitted signal to reach the far end.

“If there is direct line-of-sight (LOS) connectivity to the distant receiver, then this is a straightforward calculation,” Carpenter said. “But if the distant receiver is not in LOS, then we have to compute the right directions to push our energy so that the maximum amount of energy lands on our intended receiver. For non-LOS communication, this can lead to some strange and interesting composite radiation patterns (beams), which in some cases may actually split the energy between multiple paths.”

Because of this beamforming, the nature of the transmitted signals from each antenna will vary over time. That means that the local nulls will move around as well. “For phased-array antennas that are electronically steerable, there will be a huge challenge to achieve spatial nulling,” said Carpenter. “The locations of transmitted energy nulls will be very difficult to locate, and will probably be moving around continually depending on the various beams that the antenna will be called upon to develop from instant to instant.”

In other words, the need to maximize power at the far end for the purposes of delivering a signal competes with the need for local nulls to cancel the transmit signal near the local receiver. “At this time (with my limited knowledge) I don’t see a way through this problem for mm-wave 5G base-station arrays,” he said.

Early Ansys customers are looking at larger arrays of antennas – 8×8 or 16×16 — with each antenna located roughly a half wavelength from its neighbor. They could operate these in full duplex mode, or use one array for transmit and one for receive, or even have a 2×2 array of arrays, where they differ in polarization.

In a cellular application where the antennas are talking directly to mobile subscribers, they can look at the uplink signal for each subscriber to calculate the pre-coder weights that will optimize the beam independently for each subscriber. This is done at the baseband level for each subscriber before the signal is mixed up to RF. And this has to be done repeatedly because subscribers may move.

Paparisto explained that it’s not the carrier frequency that you have to worry about. It’s the data bandwidth itself, which will be much lower. Right now, such rates are in the 80 to 100 MHz range. With 5G, they move up to 1 to 2 GHz. He noted that in the commercial world you’re limited by standards and interop needs, so an individual designer doesn’t have the flexibility to play with signal formats. “But if you’re in the military/aerospace world, you may have more flexibility,”

This can be particularly true with spread-spectrum techniques, noted Paparisto. Here, a data stream is assigned a spreading code at the near end that divides the data across many different frequencies, each at low power. The receiver at the far end of the channel will know the spreading code and can retrieve the signal. But the return signal from the far end will be coded using a different spreading code. That means that the near-end receiver will receive it using that code. So the transmitted and received signals will very likely have power contributions from different frequencies, making it easier to separate them.

Paparisto said there are short links today that can use full duplex, like from one building to another on a campus. But these are usually custom links. Full duplex has not yet been deployed for general-purpose links of longer distances. Kumu Networks would appear to be an exception to this.

Tough chips to design
Critical for all of this to work, however, are RF ICs that can work at these elevated frequencies. Such RF chips are hard enough to design, due to the need to manage leakage effects, stray fields, and grounding issues. It becomes even harder to do cancellation for full duplex. “That requires real precision,” said Carpenter.

Most RF ICs will use the 45 and 60nm nodes, and they need III-V materials. “[Linear] components are therefore huge, especially inductors,” says Carpenter. “It’s hard to stop the leaking and coupling; it’s hard to tune to a sharp Q.”

Carpenter noted a number of remaining challenges before full duplex becomes a mainstream reality:

• Many of these DSP techniques are most effective on narrow bands. Work remains to get them to work for wide bands.
• The MIMO techniques require more work.
• It will take work to get this into a few-chip integration – say, an RF chip plus a digital chip built on an aggressive CMOS process node.
• While there is attention on 5G as a big driver, these techniques can also be applied to multiple bands and protocols, like WiFi, Bluetooth, and ZigBee.
• As new capabilities roll out through several phases, it will be a challenge to keep the spectrum from getting overcrowded.

Paparisto agreed there’s a lot of work to be done. When asked for his guess as to when this might come about, he described the situation as being driven by standards. “If providers find that they’ve used all of the frequency as much as they can [using techniques available today], then they’ll likely drive for [full duplex].” That means that they’ll approach the standards groups, which will start negotiating how this might be added to the standard. That will require a move beyond pure research to see what’s possible and reasonable to include in the spec. Once the standard is updated, then commercial offerings can appear.

But this likely will be a phased rollout. Early phones would still have to be supported. Today, frequencies are reserved for FDD and for TDD. Some would need to be reserved for full duplex when that was available. Once FDD and TDD were no longer utilized by the community of phones in use, then the FDD and TDD spectra could be reclaimed for full duplex. Paparisto believes there will be pressure in this direction in the coming years.