Practical Techniques for Measuring Interference in
Next-Generation Handsets
By Wayne Smith, Wireless Applications Marketing Engineer and Tim Masson, Senior Application Engineer and Technical Consultant, Agilent Technologies

Abstract
Interference is bound to occur when wireless technologies such as WiFi or Bluetooth™ are integrated, along with cellular functions, into the small volume of a handset. This article provides practical techniques for characterizing and measuring interference, as well as managing its impact on the system. Specific items to be discussed include the mechanisms by which circuits interfere with each other, acceptable interference levels, and two powerful measurement techniques: noise subtraction and cross power spectrum averaging.

In many ways, cellular handsets have become the electronic equivalent of the classic circus clown car. How can so many features fit into such a small package? What started as mobile voice communications has grown into a plethora of useful applications and gadgets. Today, handsets contain digital cameras, game players, MP3 music players, FM radios, internet connectivity, video broadcast services, and more. Internet connectivity and video broadcast both require high data rates with corresponding wide bandwidths. 3G has responded to this need with HSDPA and 1xEV-DO, but a new generation of more specialized technologies like WiMAX, WiFi, DVB-H in Europe, ISDB-T in Japan, and S-DMB in Korea are being integrated into handsets as a much more economical alternative.

When wireless technologies, such as Bluetooth™ and WiFi, are integrated along with cellular functions into the small volume of a handset, interference between the radios is bound to occur. This convergence of technologies is the primary reason why today’s test engineers must now measure extremely low-level interference signals. While previously it would have been impossible to make such measurements, instrumentation and techniques now exist which make measurement of these signals plausible.

Sources of Interference
Interference is problematic only when services operate concurrently. It is possible to multiplex various wireless services when they are all TDMA, but that may not be practical in all cases. Regardless, TDMA wireless services must operate concurrently with CDMA systems. For example, Bluetooth™ must operate concurrently with CDMA cellular handset operation, and CDMA and WiFi services must operate concurrently during handoff.

Before discussing how to measure interference, it is important to gain a clear understanding of the phenomenon.

Three main sources of interference can arise within the small confines of a handset:

Crosstalk via power supplies and grounding. Because space considerations make it impractical to have more than one battery power distribution system in a phone, wireless services in a handset must share a common power source. Some coupling of signals is inevitable. Here, the parasitic impedance consists mainly of the battery resistance and any series resistance in the power rail and ground that occurs before these buses are split. This coupling effect can be reduced by making the common impedance as small as possible via splitting of the current paths as early as possible, as well as through careful decoupling of the power buses.

Parasitic coupling between circuits. Given the small handset volume and the resulting close proximity of its circuits, parasitic coupling is a likely source of interference. The three main considerations for minimizing this coupling include component and circuit location, component or trace orientation, and shielding. While locating potentially interfering circuits as far apart as possible is advisable, there are many constraints on the handset physical design, including its size, which makes this an impractical solution. Some level of rejection can be gained by proper orientation of various circuits, which causes the interference to become a common mode signal. Proper trace orientation is also critical for minimizing interference. If separation and orientation are not sufficient, as is always the case in handsets, then shielding must be used.

Poor isolation between antennas. When wireless systems are colocated within a handset, coupling between antennas must be managed carefully. Meaningful physical separation (i.e., attenuation due to distance) is not practical since most phones are just 9 or 10 cm in length. That distance is meaningless for systems meant to communicate over kilometers or, in the case of Bluetooth™, just meters.

Orthogonal polarization of antennas and locating the antenna for one service in the null of the antenna for another service are techniques for reducing this interference. Circuit shields and ground planes can also shield antennas located in different sections of the phone. In cases where these techniques are not sufficient, additional filtering must be employed in either the transmitter emitting the interference or in the receiver being desensitized, depending on the exact nature of the problem. Practical levels of isolation are 20 dB for two antennas operating in the same band and potentially more than 30 dB for the cross-band case.

Interference Levels
Prior to measuring the low-level interference signals that occur from colocation of multiple wireless services in a handset, it is useful to determine the maximum allowable interference.

Out-of-band emissions, such as phase noise and spurious emissions inadvertently emitted by a transmitter often fall in the receive band of other services, raising the noise floor and desensitizing the receiver. Even relatively low level interfering signals can be significant since cell phones must operate at the peripheries of cells. Any desensitization reduces the maximum cell size and must be compensated for in the system by higher base-station transmit levels. The cumulative effect of many such phones is a reduction in system capacity. As a consequence, receivers must be designed to meet minimum system requirements in the presence of such interference.

Calculating the required isolation between a Bluetooth™ or WiFi transmitter and a GSM receiver is fairly straightforward. The required isolation is simply the difference between the maximum legal out-of-band emissions from the Bluetooth™ or WiFi transmitter and the maximum tolerable level of interference at the input of the GSM receiver. Both Bluetooth™ and WiFi operate in the ISM bands and both also defer to the ISM specifications for the country in which they will operate. The United Kingdom’s out-of-band emissions specification, ETS 300 328, states that for frequencies below 1 GHz, the maximum power of the out-of-band emissions in any 100 kHz bandwidth must be less than –36 dBm. For this example, the >–36 dBm specification is normalized to >–33 dBm in a 200 kHz bandwidth for direct comparison with GSM signals. This is the maximum interference level that can be legally emitted from the transmitter.

GSM specifications require handsets to provide no worse than a 0.1 percent bit error rate (BER) with an input signal of –102 dBm. The minimum Carrier-to-Interference (C/I) required to meet this sensitivity is 9 dB, including a 2 dB implementation margin. This means that the level for all sources of noise and interference must be less than –111 dBm to meet the sensitivity specification. In this system, the three sources of noise and interference are thermal noise, receiver added noise expressed as noise figure (NF), and received interference. The thermal noise is fixed by nature, while the receiver NF is fixed by design. Therefore, if the receiver NF is known, it is possible to calculate the maximum allowable level of interference.1 For a receiver NF of 8 dB, the interference must be >–123 dBm. Given these conditions, the required isolation between a Bluetooth™ or WiFi transmitter emitting the maximum legal interference signal power and a GSM receiver is 90 dB.

As shown in Figure 1, the maximum tolerable interference level varies as a function of the receiver NF. If the receiver NF is 10 dB, then the thermal noise plus the NF is –111 dBm, leaving no room for outside interference. Such a receiver would barely meet specifications even if there were no interference. Consequently, the NF of a receiver must be reduced to accommodate the interference signals. If a NF of 8 dB is assumed in Figure 1, the maximum tolerable interference level is -123 dBm, which agrees with the interference level calculated above.

Once the acceptable level of out-of-band emissions is known, the actual level can be measured to verify compliance or to uncover a problem. Figure 2 illustrates the instrumentation configuration for measuring interference using the following equipment:

• A spectrum analyzer, such as the Agilent E4443A PSA, to measure the emission level. The PSA’s internal noise, or DANL, is a low –153 dBm/Hz between 2 and 3 GHz, which results in a NF around 23 dB; low for a spectrum analyzer.

• A low-pass filter to block the fundamental transmitter frequency at 2.4 GHz, which keeps the transmitter from desensitizing the spectrum analyzer.

• A circulator, placed between the Device Under Test (DUT) and the filter, to keep the DUT properly terminated and operating normally when the transmitter signal is blocked or reflected by the low pass filter.

• A low noise amplifier (LNA) to reduce the NF of the analyzer from a value between 23 and 25 dB down to 7 or 8 dB.

The resolution bandwidth (RBW) of the spectrum analyzer should be set to 100 kHz since the ISM specifications for out-of-band emissions specify the level in a 100 kHz bandwidth. While the signal to noise ratio may be quite poor, using a narrower RBW will not help. With wideband signals like Bluetooth™ and WiFi, the signal power is reduced as much as the noise power when the RBW is reduced. As a result, the analyzer noise must be reduced somehow when measuring extremely low out-of-band emissions.

As an example, consider the measurement of the out-of-band emissions from a WiFi module to be used in a CDMA2000 phone. For the sake of this example, assume the power level of the interference is –109 dBm in a 100 kHz bandwidth. (This is a reasonable level if the NF of the CDMA2000 receiver is 8 or lower and the isolation between antennas is a least 25 dB). The thermal noise at the input of the LNA is –124 dBm in a 100 kHz bandwidth. However, since the NF of the LNA is 7 dB, another 7 dB of noise is added to the input noise, making the total noise –117 dBm. This results in a measured Signal-to-Noise Ratio (SNR) of 8 dB, which corresponds to a measurement error well under one dB; more than acceptable for these purposes.2

In calculating the maximum allowable interference, it was assumed that all the interference came from out-of-band emissions of the WiFi or Bluetooth™ modules. In reality, the interference from all simultaneous operating systems must be considered when trading off receiver NF and external interference levels. The effects of receiver susceptibility, such as blocking, must also be considered. Once all interference sources have been considered, and allowances made for receiver desensitization due to susceptibility problems, the allowable level from any one source may be much lower than previously calculated. This can result in much tougher design and measurement requirements.

Measuring Interference
Two techniques for measuring extremely low-level interference signals are noise subtraction and cross power spectrum averaging. These techniques are generally applicable to any wireless technology.

Noise Subtraction
Noise subtraction is an easy technique conceptually. First the signal-plus-noise is measured and averaged extensively to reduce the variance in the noise power. Next, the signal is disconnected from the analyzer, and the analyzer input is terminated. The noise is then measured and averaged separately. Finally, the average noise is subtracted from the average signal-plus-noise, and the signal is displayed and stored. A successful measurement using the noise subtraction technique is illustrated in Figure 3. Given a signal-plus-noise value of –114.7 dBm, and a signal of –127.9 dBm, the average signal level is approximately –13.2 dB below the noise level.

There are a few considerations that must be taken into account when utilizing this technique. To begin with, the spectrum analyzer must be configured for maximum sensitivity and proper detector type: set the attenuator to 0 dB, turn on the preamplifier and select average/power as the detector/average type combination. Also, the power must be averaged in watts, as opposed to dB, dBm (the log of the signal) or volts. (Note that taking the log of an averaged linear signal and averaging the log of linear signals are different operations). Having a power (true RMS) detector makes the task easier, as a true RMS detector works equally well for both broadband and narrowband signals. Additionally, the analyzer used for this technique must be stable over the relevant time and temperature ranges. For best results, measure and average over a relatively narrow frequency span.

To get the last measure of performance from noise subtraction, it is essential to replace the signal source with a termination equal (ideally) to the source output impedance when measuring and averaging the noise. Substitution of the termination allows the total thermal noise developed at the input of the analyzer to maintain the same value it had during the signal-plus-noise measurements. Since the thermal noise at the input is relatively small compared to the internal noise of the pre-amplifier, it is likely that the noise subtraction technique will work to some extent without the termination, however, this termination is critical to get the maximum possible improvement. It may also be possible to simply turn off the source signal and leave it connected, depending on the source’s characteristics. This approach will work only if neither the output impedance of the source nor its thermal noise level changes with signal level.

The key to the noise subtraction technique is averaging. Ideally, the signal-plus-noise and noise levels would be captured with a single measurement each and then subtracted to extract the signal. Unfortunately, both noise and signal-plus-noise signals have considerable variance around their average levels. By taking only one measurement of each signal, the likelihood of arriving at an unacceptable answer is exceedingly high.

The effect of using too few averages is that the variance of the extracted signal is too high. As shown in Figure 4, the probability curves (No Averaging) for the noise and signal-plus-noise signals are spread over wide power ranges. Individual readings occur randomly over these ranges in accord with the probability curves. In the first set of graphs, the Sn+Nn signal happens to be on the low side, while the corresponding Nn measurement is on the high side. The resulting Sn signal shows a nonsensical negative power. Averaging must be used to reduce the variance of the noise and signal-plus-noise signals before attempting to extract the desired low-level signal.

Figure 5 illustrates that the sigma of an averaged, uncorrelated signal is inversely proportional to the square root of the number of averages. Since variance equals sigma squared, the variance of an averaged signal is inversely proportional to the number of averages. The signals to be averaged must be uncorrelated for their variance to decrease. This is usually not a problem for the analyzer noise signal. However, the signal-plus-noise contains the signal component which is probably pseudo-random at best. Luckily, the noise component of the signal-plus-noise signal is uncorrelated and therefore will be reduced as expected. Since the variance of the signal-plus-noise signal is not reduced inversely as a function of the number of averages, it is best to base the calculation of the number of averages required on the required improvement in the variance of noise signal measurement.

In practice, noise subtraction will usually provide about 10 dB of improvement. Improvement beyond this is possible, but it asymptotically approaches some limit that depends on the nature of the signal and the characteristics of the spectrum analyzer.

The price paid for the sensitivity gained is measurement time. A 10 dB improvement in signal to noise, for example, requires a 400 fold3 increase in the measurement time. While in R&D and design verification environments the lengthy measurement time is often not all that objectionable, such an increase is rarely acceptable in manufacturing. Note the actual measurement time per averaged measurement can often be significantly improved by slowing the sweep speed of the spectrum analyzer (see Slow the Sweep Speed for Faster Measurements).

Cross Power Spectrum Averaging
Of the two measurement techniques, cross power spectrum averaging is the least known, but most powerful. As illustrated in Figure 6, this technique requires an analyzer with two independent measurement channels, with respect to their internal noise sources, that are sampled synchronously. Both channels must be connected to the same signal source with cables of the same length. The signals from the DUT in the two channels are identical and therefore highly correlated. In contrast, the noise in one channel is highly uncorrelated with the noise in the other channel because the two channels are independent.

The two spectrums are measured independently and the cross power spectrum average is calculated as shown below:

The cross power spectrum equals the signal in one channel times the conjugate of the signal in the other channel. The signal in each channel consists of the signal from the DUT being measured plus the independent noise in each respective channel. Four terms result from this calculation. The first, SS* is the power of the signal from the DUT. All of the other terms are products of signals that include at least one noise term with a mean value of zero and another uncorrelated noise or signal term. As a result, these terms average to zero.

As an example of the power of this technique, consider the measurement results illustrated in Figure 7. In this measurement, a low-level WiFi signal is extracted from the analyzer noise floor. The upper trace is a single measurement with no averaging. It has an average power of –110.3 dBm. The lower trace is the result of 51,635 averages. It has an average value of –127.2 dBm, which is an improvement of about 17 dB.

Compared to noise subtraction, cross power spectrum averaging is a much more powerful technique. Improvements above 15 dB are regularly obtained with no concern about analyzer drift over temperature and time. These results can be further improved by using low noise pre-amplifiers on each channel.

Conclusion
Interference can be a serious design problem, given the decreasing volume of next-generation handsets. Characterizing a problem is usually the first step toward solving it. Techniques like noise subtraction and cross power spectrum averaging now enable test engineers to measure the low-level interference signals that arise from the colocation of different wireless services within the handset. Agilent Technologies offers a broad range of solutions, such as the PSA series of spectrum analyzers and the VSA series of signal analyzers, to assist with these measurement techniques.

About the Authors
Wayne Smith is focused on cellular technologies as a Wireless Applications Marketing Engineer. He has worked for Agilent (and Hewlett-Packard) for 28 years in a variety of product planning and marketing positions. For the last ten years, that work has been exclusively focused in the area of RF & microwave products for the Cellular and Aerospace/Defense market segments. Wayne graduated from the University of Nebraska in Lincoln with a Bachelor of Science in Education in 1975 and from Southeast Community College in Milford, Nebraska with an Associate of Applied Arts Degree in Electronics Technology in 1966.

Tim Masson is an Application Engineer based in Agilent’s UK sales region HQ in Winnersh, Berkshire. He joined HP at the South Queensferry (Scotland) facility in 1978 and moved into application engineering in the early 1980s. For the past 20 years, he has been almost entirely focused on the design, implementation and support of test systems for cellular telephony and other wireless technologies. Tim graduated with a BSc in Physics from Nottingham University in 1971.

Reference
Agilent Spectrum Analyzer Measurements and Noise, Application Note 1303
Application Note, Literature Number 5966-4008E, December 16, 2006, Agilent Technologies, www.agilent.com/find/bluetooth

Endnotes
1 Max Interference for a GSM Receiver (dBm) = 10*LOG10 (LOG10-1(--102--9--NF) –(LOG10-1(--121))). Note: --121 dBm is the level of the thermal noise in a 200k kHz bandwidth.
2 This calculation assumes that the gain of the LNA is high enough to make the SA NF irrelevant.
3 10 times for each 5 dB, times 2 because the signal-plus-noise and the noise must be measured separately, and times 2 again because subtracting the averaged noise from the averaged signal-plus-noise increases the sigma of the difference signal by the square root of 2.

AGILENT TECHNOLOGIES
www.home.agilent.com
TXTLINX.COM82
Email this article to a friend!