Comparing Analog and Digital Techniques for Measuring Noise Power Ratio

Generally, noise power ratio (NPR) measurement is accomplished with a diode-based noise generator as a stimulus and a spectrum analyzer for analysis. However, a digitally generated stimulus and vector signal analyzer deliver more repeatable results in less time.

Roland Hassun
Hewlett-Packard Co.,
Santa Rosa Systems Division
Santa Rosa, CA

In NPR measurements, an analog or digital stimulus is applied to a device under test (DUT) and the device's output is analyzed by analog or digital means. Of the four analog and digital measurement configurations, the analog stimulus/analog analysis method is the most common and relies on a noise source with white Gaussian characteristics as the stimulus and a spectrum analyzer for analysis. The digital input/digital analysis method utilizes a digital waveform generator to provide a synthesized discrete stimulus signal and a vector signal analyzer for determining NPR.

Both measurement techniques deliver acceptable estimates of NPR, and diode-based noise sources are certainly more familiar. However, the digital stimulus/digital analysis method reduces NPR test time by an order of magnitude and is more repeatable from test station to test station. The benefits make this method a good choice when used in high volume manufacturing operations.

Three Decades New

NPR was first conceived more than 30 years ago to provide an objective, quantitative way to characterize the intermodulation distortion (IMD) of active circuits in telecommunications systems. IMD occurs when two or more signals mix in a nonlinear device such as a PIN diode, producing spectral lines in addition to the input signal, as shown in Figure 1 . These additional signals degrade the bit error rate performance of communications systems significantly and cause interference with services operating on adjacent channels. The goal of NPR measurement is to characterize fully the performance of a device or subsystem in terms of its IMD.

Fig. 1: Additional spectral lines from IMD.

NPR is not an ideal figure of merit because it is dependent on the characteristics of the stimulus and it is sensitive to the measurement technique employed. NPR measurement also requires the use of a calibrated stimulus (rather than a simple sinusoidal signal) that closely approximates the characteristics of actual signal traffic. However, when conducted properly with either analog or digital equipment, NPR is a valuable measurement tool for the communications system designer.

Natural vs. Synthetic Noise

An analog stimulus utilized to evaluate NPR usually is generated by an avalanche diode that, when subjected to a voltage, produces a spectrum of natural noise with a white Gaussian distribution from which a portion of the noise spectrum (the notch) is removed, as shown in Figure 2 . The noise does not repeat in time and therefore has a continuous spectral distribution.

Fig. 2: A typical noise spectrum generated by an avalanche diode.

The natural noise statistics depend on the characteristics of the diode, and their average power is constant only when measured over a long period. In addition, peaks in amplitude occur at random so they are not predictable. The analog method provides uniform distortion components in the notch. In addition, the technique is well understood since it was employed universally throughout the RF and microwave industry until the introduction of digital waveform generators and digital test equipment.

On the other hand, the digital method is relatively new and its attributes are not generally well known. As its name implies, synthetic noise is created digitally by a waveform generator, such as the multiformat communications signal simulator (MCSS) used to obtain the results described. The period of synthetic noise is predictable and its statistics are well defined since they are created synthetically.

A synthetic stimulus allows triggers and timing signals to be used to start and end the averaging period via markers that trigger the analysis instrument every time a particular waveform appears. The triggering ability is useful when IMD products are masked by device or thermal noise, reducing the signal-to-noise ratio to a point at which the desired signals cannot be distinguished from the background. This situation arises when testing satellite transponders, which have extremely high gain. By triggering, measuring and averaging only on the synthetic waveform, signal-to-noise ratio can be improved significantly.

With either the analog or digital technique, the signal is applied to the input of the DUT and the resulting output is observed. Any nonlinearity in the DUT produces spectral components within the notch, as shown in Figure 3 . NPR is simply the ratio of the power in the pass band to that of the spectral components in the notch.

Fig. 3: Spectral components within the notch produced by nonlinearity.

Analog Technique

In this configuration, the input stimulus is analog (an avalanche diode) as is the output measurement tool (a swept spectrum analyzer). This natural noise stimulus is conditioned according to the requirements of the test with amplification, frequency translation, and bandpass and band-rejection filtering. The signal spectrum of the input stimulus depends on the shape of the filters employed. Figure 4 shows the effects of filtering on a spectrum produced by an analog noise source. Repeatability from unit to unit depends on the filters and generally is rather poor.

Fig. 4: The analog noise source's spectrum as determined by filtering; (a) broadband and (b) narrowband.

The resolution bandwidth of the spectrum analyzer is set as wide as the instrument allows while maintaining its ability to resolve the notch accurately. For example, if there is a 25 MHz noise bandwidth and 200 kHz notch, the resolution bandwidth should be no more than 10 kHz since the clean, rectangular shape of the notch will smooth out eventually as the bandwidth is increased.

When natural noise is passed through a filter that is 10 kHz wide and then sampled, the spacing of the samples must exceed 1/(filter bandwidth), which in this case is 100 micron, for the samples to be independent statistically. To obtain estimates of the RMS values that have standard deviations of 0.1, the averaging time must be at least 10 ms (100 × 100 micron).

Since the power of a diode-based noise source fluctuates with time, the longer the averaging period, the more accurate and repeatable the measured values will be. The standard deviation of the measured power is 1/(square root of the number of samples). If several sets of 100 samples are taken and the RMS value of each set is computed, the standard deviation of the RMS power of each set is 0.1 or approximately ±0.5 dB.

A spectrum analyzer sweeping over a 1 MHz band with a 10 kHz resolution bandwidth requires a video filter to slow down the sweep to at least 1 MHz/10 kHz × 10 micron = 1 s. This condition requires that the standard deviation of the fluctuations be less than 1 dB (approximately 100 mV) and that 96 percent of the fluctuations are within 3 dB (±1.5 dB). This measure of repeatability can be improved to a more acceptable ±0.5 dB by taking 10 averages of the one-second sweep. However, further improvements to ±0.25 dB require 40 averages and a measurement time of 40 seconds. When considered in terms of a high speed production line, this measurement time is lengthy.

The Digital Technique

A digital stimulus/digital analysis NPR measurement system, shown in Figure 5 , is similar to those utilized for evaluating most of the parameters of communications link quality, including gain, gain flatness, group delay, noise figure, frequency stability and error vector magnitude. The stimulus side of the system includes a source of complex waveforms (the MCSS) and an upconverter. The signal is reproduced by the MCSS with 12-bit resolution and sent to the DUT. The output of the DUT is fed to a wideband receiver and then to the vector signal analyzer for NPR calculation. The instrument allows two 50 kHz band segments to be evaluated within the notch, producing an effective measurement bandwidth of 100 kHz. The resolution bandwidth is set to 10 kHz to ensure that the notch is resolved clearly. Acquisition time required with these settings is approximately 100 micron.

Fig. 5: A digital NPR measurement system.

The separation of spectral lines generated by the stimulus can be 238, 476 or 952 Hz, and is set by the MCSS. These separations provide 419, 209 and 104 spectral lines, respectively. The number of spectral lines in the digital method is somewhat similar to the number of samples gathered in the analog method. The actual number of spectral lines has proven to have little impact on the accuracy of the measurement, as listed in Table 1 .

Table I: NPR as a Function of Stimulus Spectral Density

Line Spacing (Hz)












The standard deviation of the measured RMS value for the stated separations is 0.41, 0.58 and 0.81, respectively, and the associated repeatability is ±0.6, 0.87 and 1.2 dB. This deviation represents the fluctuation of each separation that can be expected under most measurement conditions.

The waveform of each DUT can be captured with a single 100 micron acquisition, but the processing time required to determine NPR is approximately two to three seconds to ensure a repeatability of ±0.2 dB. As a result, the entire measurement takes approximately three to four seconds to perform, compared to up to 40 seconds for the analog/analog method. Repeatability improvements are obtained by averaging measurements; 10 averages provide a factor of three improvement in repeatability. This averaging makes the measurement effective in the presence of additive noise or random phenomenon such as jitter.

In Practice

The use of an analog white noise stimulus and spectrum analyzer for evaluating NPR long has been a verifiable tool for analyzing the IMD performance of communications components and systems. However, use of an analog stimulus on the production line results in measurement times in the tens of seconds, which over even a single day's production run consume a large amount of time. By contrast, a synthetic stimulus and digital signal analyzer for analysis require an order of magnitude less time per measurement and deliver equivalent results (or better) with greater repeatability. The shape of an analog stimulus is also comparatively less well defined than that of a synthetic stimulus, which places additional stress on a DUT even if the average powers of the two stimuli are identical.

An alternative to the analog/analog method is to substitute a digital waveform generator for the noise source. In this case, spectral separation is defined as 1/(duration of acquisition) and the magnitude of the spectral lines varies considerably. To achieve spectral uniformity within the notch, more spectral lines can be placed within the measurement bandwidth of the spectrum analyzer. Unfortunately, the greater amount of information that must be handled by the waveform generator may exceed the capacity of its internal memory. The additional processing also requires much more time to achieve the desired level of spectral uniformity.


The continuous spectrum of the analog stimulus with its nearly uniform amplitude in the notch is familiar and comforting, but comes at the expense of the long observation time required for averaging. This measurement time may not be apparent when making manual measurements and, in fact, may not matter when making a presentation or giving a lecture. However, it is an imposing limitation in the production environment where throughput is related directly to revenue and, hence, to profit.

While both analog and digital stimuli and analyses are equivalent in their end result, they achieve it differently. In the case of the digital alternative, this end result is achieved more quickly with more measurement uniformity from unit to unit. As a result, the synthetic stimulus is a good choice when NPR must be measured on the production line or in any application where measurement speed is a consideration.


The MCSS used is the model HP E2507E manufactured by Hewlett-Packard Co., Santa Rosa, CA.