We are experiencing continuing and substantial growth in the IEEE 802.11 WLAN marketplace. Much of this growth is being fueled by the introduction of dual-band radios. These radios are capable of operating under the IEEE 802.11a protocol at 5 GHz and the IEEE 802.11b and (proposed) 802.11g protocols at 2.45 GHz. They can operate at data rates that range from the legacy 1 and 2 Mbps to the state-of-the-art 54 Mbps. As a result, dual-band radios are capable of ubiquitously communicating with all known 802.11 access points.
Though this capability should make dual-band radios attractive to designers, their optimization, testing, and debugging poses some unique problems. For example, the high data rates use complex 64-QAM modulation, which requires special test equipment and techniques for troubleshooting. This modulation requires a high degree of linearity in both the receiver and transmitter in order to prevent excessive error vector magnitude (EVM). The front-end circuitry also is a factor. It now contains separate 2-GHz and 5-GHz RF signal paths in addition to common paths. Because of the large assortment of possible data rates, unique issues emerge concerning the minimization of test time in high-volume applications.
A dual-band 802.11a/802.11g radio has nine unique modulation modes. Each of these modulation modes requires a different closed-loop power-control setting for optimum performance:
- 2-GHz CCK modes (BPSK or QPSK)
- 2-GHz 6- and 9-Mbps OFDM (BPSK)
- 2-GHz 12- and 18-Mbps OFDM (QPSK)
- 2-GHz 24- and 36-Mbps OFDM (16 QAM)
- 2-GHz 48- and 54-Mbps OFDM (64 QAM)
- 5-GHz 6- and 9-Mbps OFDM (BPSK)
- 5-GHz 12- and 18-Mbps OFDM (QPSK)
- 5-GHz 24- and 36-Mbps OFDM (16 QAM)
- 5-GHz 48- and 54-Mbps OFDM (64 QAM)
In addition, these settings can vary somewhat over the frequency band. It is therefore prudent to provide the capability of storing this data for multiple channels. Practically, it is usually adequate to store this data at three channels: one at the low end of the band, one in the center, and one at the high end. The designer can then use a best-fit curve to interpolate the data for other channels.
Fortunately, the 802.11b CCK modes are a bit less complicated to test than the OFDM modes. Taking into account this consideration, however, it remains clear that the dual-band network-interface-card (NIC) radio is approximately 2.5 times more complicated to test than a single-band radio. As a result, it is a considerable challenge to design an automatic-test-equipment (ATE) suite that yields reasonable execution time (say 2.5 to 3 min. per radio).
Another considerable challenge is the measurement of error vector magnitude. Take the case of the 54-Mbps data rate with 64-QAM modulation. In this instance, the 802.11 specifications require that a minimum EVM level of −25 dB be maintained in the transmitter. This requirement is quite an important one. In a typical system, documented evidence shows some degradation of receiver sensitivity if this transmitter EVM level is not maintained. This therefore becomes a potential interoperability issue. In order to avoid system-range issues, strict adherence to this EVM specification is mandatory.
Competent laboratory-grade vector signal analyzers (VSAs) are available that can measure EVM. But they are high-ticket items costing in excess of $80,000. While this cost is perhaps acceptable in an engineering-laboratory situation, it is certainly not desirable in a production test environment. Here, more than 20 test stations may be needed to ship the necessary volume of NICs.
Fortunately, the major test-equipment suppliers and the NIC manufacturers are working to address this issue through an ongoing dialog. They are being prompted by the already large volume of 802.11x NICs currently being manufactured, as well as those planned for the near future. Hopefully, a breakthrough in the price point will emerge in the medium term. Certainly, complex modulation formats such as 64-QAM are here to stay. In fact, they are destined to grow in usage.
Until these production-test issues are resolved, however, engineers will have to take innovative approaches. For example, it is possible to avoid testing a large percentage of NICs for EVM in production by first correlating the EVM with the spectral mask. In this method, a reasonably large sample of NICs is tested. The NICs are then used to correlate 54-Mbps EVM readings with the resultant transmitter spectral mask. Using this data, the actual production radios are aligned based on the spectral-mask measurement. Hopefully, a small sample from each lot also will be re-checked for EVM. This step will ensure that the correlation is holding constant and that the variance in these two measurements is low.
The advantage of this approach is that a substantially lower-cost instrument—namely a spectrum analyzer—may be used for high-volume testing. This same instrument is needed anyway. It is already being used for the proper alignment of lower-data-rate constellations, so it incurs no incremental cost.
Of course, a disadvantage exists with this method as well. To guarantee EVM compliance, the 54-Mbps power output must inherently be set a bit on the conservative side. A given NIC may therefore produce slightly lower RF power output than it would if it was really aligned for target EVM. Typically, however, the sacrifice in power output attained by this method will only be around 1 dB. Because the indoor range of a WLAN varies as approximately the fourth root of the power output\[2\], the falloff in range would be approximately 6%. That is not a bad tradeoff.
In a dual-band WLAN, it is less expensive to fabricate both the 2-GHz and 5-GHz power amplifiers (PAs) on one piece of substrate. Yet this cost-saving decision can create a unique problem. The output of the 2-GHz PA is generally rich in harmonic energy. Yet this output can propagate to the 5-GHz output pin of the PA via coupling either internal or external to the device. Once this occurs, there is a direct, low-loss path through the 5-GHz diplexer port right to the dual-band antennas. Needless to say, the dual-band antennas are designed to be efficient radiators at 5 GHz—approximately the second harmonic of the 2-GHz signal. Typical coupling is such that the second-harmonic level is often too high to pass FCC and ETSI regulatory compliance.
Interestingly, the most critical case occurs in the 802.11b 1-Mbps legacy BPSK mode. Due to the modulation phase being either 0° or 180°, the second harmonic correlates to one spot frequency. In other words, the second harmonic of a 180° phase shift is itself 360°. Recall that this fact is commonly used for carrier recovery in such a BPSK system. In other 802.11a and g modes, the harmonic energy is more spread out in frequency. As a result, it is seldom a compliance issue.
A solution to this problem is to place a second-harmonic trap directly at the 2-GHz PA output (see figure). The circuit is series resonant at the second harmonic of the 2.45-GHz signal, so it effectively shunts this energy to ground. Being inductive at 2.45 GHz, the circuit forms part of the matching circuit at the PA output. At the same time, it provides a DC path to feed the Vcc voltage into the power amplifier's output collector circuit. (A patent is pending for this circuit application.)
If space permits, it is also possible to realize this trap by utilizing a short-circuited transmission line which is 1/2 wavelength long at the second-harmonic frequency\[1\]. At the fundamental frequency, the line is 1/4 wavelength long. It therefore appears to be open circuited. Because most contemporary designs are quite compact, however, the lumped circuit trap illustrated in the figure is usually preferred.
It was mentioned earlier that dual-band radios support nine unique modulation modes: five at 2.45 GHz (IEEE 802.11g) and four at 5 GHz (IEEE 802.11a). As the modulation complexity increases, the RF power peak-to-average ratio (PAR) also rises dramatically. In the IEEE 802.11b CCK mode at 2.45 GHz, for example, the PAR is approximately 2.5 dB. Yet with the IEEE 802.11a or 802.11g 64-QAM OFDM modulation used at 48- and 54-Mbps data rates, the PAR can exceed 13 dB.
It follows that a given PA must be considerably backed off in power when running 64 QAM as compared to simpler modulation modes, such as CCK or BPSK OFDM. To quantify this, a typical radio may run an average RF power output of 18 to 19 dBm in CCK modes but only 12 to 13 dBm in 64 QAM. Going to the higher data rate results in the loss of roughly 6 dB of power output.
Now look at the receiver side of the equation. For a given packet error rate (PER) of 8% to 10%, 11-Mbps CCK theoretically requires a symbol-to-noise ratio (Es/No) of 18 dB (assuming 2 dB of system implementation loss). Yet a 54-Mbps 64 QAM requires 26 dB, so another 8 dB has been lost here.
Summing up the transmitter and receiver losses, a whopping 14 dB has been lost just by going from the 11-Mbps CCK to the 54-Mbps OFDM data rates. As a link power ratio, that equals (10)1.4 = 25:1. In an indoor, crowded office environment, the range falloff is proportional to somewhere between the cube root and fourth root of the power ratio\[2\]. As a result, the expectation here is that at the same frequency, the range of 11-Mbps CCK will be between 2.2 and 2.9 times greater than that of 54-Mbps OFDM. Data from actual hallway measurements agrees fairly well with this prediction. The conclusion is that all of the performance possible must be squeezed out of the radio.
Needless to say, the transmitter's power output must be meticulously optimized while preserving the necessary 54-Mbps EVM (−25 dB in the case of IEEE 802.11a and g). The same can be said for minimizing the receiver's noise figure. It also holds true when ensuring that all front-end losses in the transmitter and receiver are as low as possible.
One possible additional area of focus is the antenna. In a practical WLAN client radio, the designer must essentially strive for an omnidirectional azimuth antenna pattern. Doing so will guarantee good performance at any random orientation angle with respect to the access point (AP). To a large extent, a practical antenna will have a gain right around 0 dBi. Keep in mind that any practical antenna used in a laptop computer will have peaks and valleys in its pattern due to reflections from the laptop itself.
The search for the best antenna pattern is further complicated when including the high-multipath environment in which a typical client radio operates. The effective polarization angle of the signal arriving at the AP is therefore quite random. In an indoor high-multipath environment in which a WLAN is typically deployed, it does not seem to matter in practice—if the native polarization of the client's antenna is predominantly vertical or horizontal. No significant difference in range has been noted when comparing these two cases. This is in contrast to a low-multipath outdoor environment, in which there seems to be some advantage to vertically polarized antennas.
Of great significance, however, is the quality of the antenna impedance matching as well as the matching of other interstage components. Many engineers deem a 2:1 antenna VSWR adequate (i.e., a return loss of 10 dB). But this would allow operation anywhere on a 2:1 VSWR circle on the Smith Chart—in the extreme case, either 25 or 100 ohms! To squeeze out that last db of performance, it's a good idea to maintain a maximum antenna VSWR of 1.5:1 (i.e., a return loss of −14 dB).
An analysis of front-end losses can be quite misleading if it is based only on reflected power. Consider that a modern front end typically contains an RF switch, diplexers, and various low-pass and band-pass filters. The insertion-loss and frequency characteristics of all of these components are dependent on each seeing the proper driving point and load impedance. Furthermore, upon inspection of datasheets for these devices, it is often evident that each of these components, in turn, may be allowed to have a VSWR of 2:1. The combination of these issues results in a significant cascaded effect of such mismatches on the overall front-end performance.
Fortunately, the input impedance of a given component will usually fall within relatively tight bounds—although not generally at 50 + j0 ohms. Most manufacturers of such components will usually provide typical S11 data. Consider placing L section matching networks between strategic front-end components in order to minimize their losses. In many cases, only a single component may be necessary to trim out the lion's share of the loss. Designing to typical specs can be dangerous, because these are not guaranteed. The best approach is therefore to negotiate with the component manufacturer to yield tighter VSWR specs.
This article has discussed several unique issues relating to IEEE 802.11a/802.11g Client NIC radios:
- Testing considerations
- 2-GHz PA second harmonic
- Power-amplifier backoff, receiver sensitivity, and range
- Optimizing front-end circuitry
The first three issues are obviously somewhat unique to dual-band OFDM radios. The discussion relating to optimizing front-end circuitry is perhaps more universal and probably applicable to many other radio designs. Designs involving high-order modulation schemes, such as 64 QAM, require extremely high-linearity PAs, low-noise amplifiers (LNAs), and mixers. After all, they need to process high PAR signals over a wide dynamic range.