The 60-GHz band has been allocated worldwide for unlicensed wireless-communications systems. The United States, Europe and Japan have all allocated at least 5 GHz of contiguous bandwidth, which is nearly equal to the total for all other wireless communications. This unprecedented amount holds the potential for much higher data rates than other channels that are bandwidth-limited, with wireless data rates in the range of 1 Gbit/second becoming reasonable.
There are some major disadvantages, however. Probably the most important one for low-power wireless communications is the path loss-for an omnidirectional antenna at 60 GHz it is severe. Another difficulty is that circuits are more difficult to design and implement compared with the lower-frequency bands in widespread use today.
In the interest of higher system integration-hence lower system cost and power consumption-we would like to implement all the circuits in standard CMOS, in this case 130 nanometers. This also compounds the problem of implementing the circuits. The fMAX of NMOS devices in 130-nm CMOS is approximately 130 GHz, which means that under optimal conditions a single-stage amplifier can be expected to achieve no more than 4 dB to 6 dB of gain at 60 GHz.
CMOS has an advantage: It can support very large amounts of digital processing in a very small area using very low power. The key to building advanced communications systems in CMOS, especially at a 60-GHz carrier frequency, is to leverage the digital computational power of CMOS.
In shifting to the higher 60-GHz carrier frequency we are also reducing the wavelength of the carrier signal, which has a marked effect on the antenna system that is required. If a nominally omnidirectional antenna is used, the path loss will scale inversely with the square of the wavelength. This means that the extra path loss incurred in moving from 5 GHz to 60 GHz is 22 dB, for a grand total of 88 dB of path loss at 10 meters for an omnidirectional antenna at a 60-GHz carrier frequency. Furthermore, it is reasonable to assume that the transmit power of a 60-GHz communications system will be roughly equal to that of its cousins operating at lower frequencies.
However, because of the limitations of CMOS circuits, it will be more difficult to achieve that power. Under these two assumptions it quickly becomes apparent that in order for a 60-GHz system to operate at high data rates at a reasonable range, the path loss must be recouped. The only way to overcome this loss is to use a high-gain antenna.
The only practical way to achieve high antenna gain in a changing or mobile environment is to use an adaptive antenna array. The key is that this class of antennas can achieve high gain in a nearly arbitrary direction.
The number of antennas for the system should be chosen to combat the path loss incurred by the system and even further improve the signal to aid the rest of the system. For rather arbitrary system specifications of 1 Gbit/s at 10 meters, with some safety margin, we can see that approximately 16 antennas will be needed on the transmit and receive sides.
In utilizing an array of antennas we reap the benefit of increased antenna directivity. However, there is another very important benefit. Now that there are N separate signals driving N separate antennas, there can be N separate power amplifiers (PAs). The total transmit power is then the sum of the array power that has been combined in space. The power output requirement for each individual PA is then reduced to 1/N of the total required for the system, where N is the number of elements in the array. This reduction in power is critical for future CMOS designs where supply voltages are near or even below 1 volt.
When adding antennas to the transmitter we obtain a double win for each antenna added: The directivity is increased by one, but the total transmit power is increased by one unit as well, if we add a duplicate PA for each new antenna. Therefore, if we add a duplicate PA for each new antenna, the effective isotropic radiated power is quadrupled for each doubling in the number of antennas. This means that a system that uses 16 antennas for the transmitter and 16 antennas for the receiver will have a system link gain of (16+16+16) = 36 dB over a system with a single antenna and a single PA.
The phases of the array should be controlled digitally; that means that the phase shifts will be quantized. In our work, simulations were carried out to test the accuracy requirements. A beam was swept from 0 degrees to 90 degrees ; at each point an ideal beam with floating-point weighting coefficients was compared with a beam formed by quantizing the weighting coefficients. The angle of the main beam was determined for the quantized case, along with the angular error compared with the ideal case. It was found that each antenna only needs a 3-bit phase shifter to achieve an angular error of a few degrees.
If the RF front end uses differential signaling and the mixers use an I/Q mixing scheme, the phase shifters will have both I and Q differential signals to work with. Therefore, the 3-bit phase shifters can be implemented as vector modulators and are then reduced to a combination of 1.5-bit circuits on the I and Q channels, called PIC circuits. These very simple circuits only need to pass, invert or cancel their input signals. The outputs of the PIC circuit on the I channel and the PIC circuit on the Q channel are then combined to form the phase shifter.
So the eight possible combinations of the I and Q PIC circuits at the output of the phase shifters are: (I), (I+Q), (Q), (-I+Q), (-I), (-I-Q), (-Q), (I-Q). The phase shifters need only switch at the channel innovation rate, which is in the range of a few hundred hertz to the low kilohertz for a 60-GHz system. If the data rate is 1 Gbit/s, it is evident that the phase shifters will appear to be static over long intervals, further easing the performance requirements that are placed on them.
A system using RF phase shifters has the advantage of having only one digital-to-analog converter, one RF up-mixer, one RF down-mixer and one analog-to-digital converter. Previous implementations used N RF mixers or even N full transceivers, which all consume more power and chip area. It also retains the advantages of having N PAs and N low-noise amplifiers.
It is important to note that the phase shifters are before the PAs in the transmitter and after the LNAs in the receiver. This means that the phase shifters in the transmitter do not need to handle high input power, nor do they need to achieve significant power gain. The situation is similar in the receiver. The phase shifters will operate on a signal that has already been amplified. Keeping this in mind, the addition of noise should be less critical, since the phase shifters will already enjoy a healthy input signal level.
The performance constraints on the RF phase shifters are far lower than what would be required for RF mixers. Most important, the switching frequency goes from the full carrier frequency of 60 GHz for an RF mixer to the channel innovation rate of a few hundred hertz or fewer for the RF phase shifters. Therefore, if the phase shifters are properly designed, the use of RF phase shifters should give all the benefits of adaptive antenna arrays, with the minimum added power, or even an overall power savings vs. more complex topologies.
Sayf Alalusi (firstname.lastname@example.org) is a graduate student in the Berkeley Wireless Research Center at the University of California at Berkeley.
See related chart