November 17, 2020
IQ capture is a technique used to effectively measure digital RF signal strength. When conducting this analysis, it is important to understand the relationship between noise, sampling rates, and bit depth. By doing so, the receiver – typically a spectrum analyzer – capturing IQ data can be optimized, making it easier and faster for engineers to identify signals in post processing.
Digital signals are “quantized,” meaning the analog signal is rounded to the nearest bit, creating error dubbed “quantization noise.” When the least significant bits are chaotically changing due to thermal noise or a wideband signal, quantization noise is “white,” or uniformly spread across all frequencies. When quantization noise has this property, it may be treated the same as any other noise source. Generally, it is preferred that the quantization noise be smaller than the analog noise.
Lowering the resolution bandwidth (RBW) on the receiver during analysis will reduce quantization noise. There are occasions, however, where quantization noise is not at all “noise-like.” If there is very little analog noise in a system and if the signal power is small compared to 1 bit, it is possible to have a situation where no bits are being toggled. In this case, the quantization noise is not noise-like; rather it is highly correlated to the signal with opposite sign.
Reducing Quantization Noise
When the quantization noise is uncorrelated to the signal, it – along with thermal noise – degrades the signal. If the quantization noise can be made much smaller than the thermal noise, then the quantization noise becomes irrelevant. It is often a design goal to make the quantization noise 10 dB less than the other noise sources. Otherwise, it will impair the signal. It is sometimes difficult, however, to make the quantization noise lower than analog noise. In this scenario, the system performance is limited by the digital performance.
While a higher sample rate reduces the quantization noise energy density, more bits are usually a more efficient approach. One additional bit reduces the quantization noise energy density by 6 dB. Compare this to doubling the sample rate, which reduces the quantization noise energy density by 3 dB. A 10-bit capture at 200 MSPS, therefore, has the same quantization noise energy density as a 12-bit capture at 25 MSPS. Engineers need to choose the minimum sample rate necessary to capture a signal that needs to be analyzed, then select a bit depth whereby digital quantization noise is lower than analog noise.
Selecting the Right Spectrum Analyzer
A spectrum analyzer has analog amplifiers and attenuators to optimize the level of the signal. To make the best use of a limited number of bits, it can be useful to fine tune receiver gain. If the input signal is too high, the signal will overflow the fixed bits. To prevent this undesirable scenario, more attenuation should be used. If the input signal is low, then the quantization noise is higher compared to the signal.
Broadband signals generally have more energy hitting the receiver’s analog-to-digital converter (ADC) than energy inside the instrument’s RBW filter. To manage this, an instrument does not overrange many dB above the reference level. In general, the best choice of bits can be made by reducing attenuation or adding amplification until the instrument overranges. Once that has been determined, users need to add 5 dB attenuation to prevent it from overranging during the signal capture.
Figure 1 is a plot of a signal captured by the Anritsu Field Master Pro™ MS2090A at a specified sample rate and bit depth. This figure shows how the choice of bit depth changes the noise floor. Near the carrier, 8 bits is sufficient, as the combined phase noise of the signal and receiver exceed the 8 bit quantization noise. By 10 MHz offset from the carrier, the 8 bit quantization noise is clearly higher than the analog noise floor and is degrading the measurement.
The 200 MSPS capture uses an analog filter with a 110 MHz wide passband. At 70 MHz offset, the receiver is no longer receiving much of the analog signal due to the analog filter. The result is the quantization noise of the lower bit depths becomes visible. The lower noise floor near the edges does not matter for actual measurements, as the signal has been filtered out at these frequencies. At this sample rate, 10 bits is usually sufficient for measurements.
With an understanding of these interdependencies, engineers can optimize the configuration of a receiver, such as the Field Master Pro, when capturing IQ data to minimize the memory size of the resulting data file. Smaller files are typically easier to analyze in post processing tools, making the identification of all signals faster and easier.
The Field Master Pro MS2090A allows captures of up to 2 GB to be stored. Choosing a lower sample rate or bandwidth will allow a longer time record to be captured. Table 1 shows how many seconds it takes to fill 2 GB. The Field Master Pro MS2090A can also stream data through a user selected Ethernet/USB3/high-speed data port. This data may be captured to a file and post processed by a PC application for detailed analysis, such as the Bird Spectro-X.
Table 2 shows recommended bit depths at various sample rates. Cells labeled “Too Much” indicate that the extra bits are highly unlikely to improve the measurement. The “Too Little” cells designate that the quantized bit depth is likely to degrade the measurement, although the measurements may still be quite usable. “Almost” cells show that the quantization noise is unlikely to significantly impact the measurement, but it might for certain signals.
To learn more about how to effectively conduct IQ capture on digital signals, you can download a white paper entitled Choosing the Best Bit Depth for IQ Captures or Streams.