Steve Sandler, Charles Hymowitz, AEI Systems Inc.
Electronic Design
Instead of your trusty old oscilloscope, try using a spectrum analyzer to measure the noise sources that degrade ADC performance
Noise from the power supply, linear regulators, and voltage references is a major contributor to the limitations of system performance, especially in instrumentation and communications products. In analog-to-digital converter (ADC) applications, the noise from regulators and references results in clock jitter, which can significantly degrade ADC characteristics such as signal-to-noise ratio (SNR), signal-to-noise and distortion (SINAD) ratio, and bit error rate (BER). Low-noise amplifiers (LNAs) also suffer from phase noise and modulation effects related to power supply noise.
Oscilloscopes often are used to measure power supply, linear regulator, and reference noise. Since an oscilloscope has a sensitivity in the range of 2 mV per division, a substantial voltage gain must be added to see the ripple and noise, which is often in microvolts. This gain is usually accomplished using a “low-noise” operational amplifier or several cascaded low-noise op amps. An active filter follows the op amps, providing high pass and low pass elements, to meet the desired measurement bandwidth with the entire circuit constructed in a Faraday shield. (A paint can serves this purpose.) Several IC manufacturers have application notes describing the measurement (Fig. 1).
 |
|
| Figure 1. | The typical measurement setup for testing noise using an oscilloscope requires a very high-gain LNA, active filter, and peak detector, making it one of the more difficult measurements. It also provides less information than other methods, since no frequency relationship is provided from this measurement. |
Several limitations are evident in this setup. First, building such a configuration takes time, effort, and extreme care. Next, the high gain required often limits the bandwidth of the measurement, and the amplifiers provide a noise path through the power supply rejection ratio (PSRR), making the circuit sensitive to the power supplies that power the circuit. In addition, the amplifier itself contributes noise.
Better Measurement Methods
The Agilent N9020A signal analyzer (with Option 503) and the Tektronix RSA5103A and RSA5106A real-time spectrum analyzers (RSAs), in conjunction with Picotest’s signal injectors, offer two ways to measure power supply, voltage regulator, and voltage reference noise. These spectrum analyzers can measure from 1 Hz to 3 GHz (RSA5103A) or 6 GHz (RSA5106A) and offer a much greater dynamic range than an oscilloscope. Both also provide an outstanding noise floor and much greater sensitivity than an oscilloscope. They provide peak detection, averaging, and high-resolution options for analyzing the results as well.
The N9020A (with Option 503) can measure noise from 20 Hz to 3.6 GHz, with other models reaching as high as 26.5 GHz. It offers many methods of acquisition and analysis options including direct spectrum measurement and oscillator phase noise and jitter. The RSA5103A and RSA5106A offer many other measurements including phase noise and jitter.
There are two basic methods for measuring voltage regulator/reference noise. One method is to measure the phase noise of a high-performance clock, which is powered by the regulator under test. A phase noise measurement of a crystal oscillator offers an effective indirect measurement of regulator noise. The noise signals from the regulator appear as amplitude modulation and as mixing products in the oscillator frequency. The phase noise measurement identifies specific noise frequencies, which can be seen as “spurs.”
 |
|
| Figure 2. | Phase noise measurements (showing the 250-kHz noise representing the power supply) show all frequency content as a mixing product in the oscillator. In this example, a single frequency is added to the power supply voltage. The resulting signal shows up in the oscillator phase noise. |
These spurs include all of the frequencies in the power supply noise, as well as all mixing products of the clock and power supply noise frequencies. All power supply noise contributes to the phase noise and can be seen in the total jitter performance, which is specified directly in the RSA phase noise display. Figure 2 shows an example of oscillator phase noise with a 250-kHz power supply noise signal. A typical power supply will result in many interference signals. Only one example is shown here for clarity. To determine the power supply noise from this phase noise plot, it is necessary to quantify the PSRR of the clock.
|
|
|
| Figure 3. | Directly connecting the DUT to the N9020A for a noise measurement allows a direct frequency-related measurement, which can be displayed as noise or as noise density. |
The second noise measurement method calls for directly measuring the device under test (DUT) (Fig. 3). To demonstrate the validity and the sensitivity of its noise measurement, the noise floor of a typical measurement setup was recorded. The noise floor is measured as a direct measurement, including the Picotest J2130A dc bias injector used as a dc blocker and the J2180A preamp. The J2180A improves the noise floor by nearly 20 dB, while also presenting a high-impedance connection to the DUT (Fig. 4). This is important since the 50-Ω termination could easily impact the noise result.
|
|
|
| Figure 4. | During a low-frequency noise floor test (with the J2180A preamp (blue trace) and without the preamp (yellow trace)), the preamp improves the noise floor by nearly 20 dB as a result of the 20-dB gain and the preamp’s very low-noise design. The J2130A dc bias injector is used to block dc so the dc does not overload the analyzer front end at the output of the preamp. |
|
|
|
| Figure 5. | Validating the N9020A noise measurement using the J2140A attenuators solely verifies that the measurement amplitude is correct by measuring a very small known signal. The J2140A attenuators reduce the signal level to a very low level (5 µVRMS). |
Next, an arbitrary waveform generator (AWG) is set to provide a 1-kHz, 50-mVRMS sine output. A pair of Picotest J2140A cascadable attenuators, each configured for an attenuation of 40 dB, is connected between the AWG and the N9020A (Fig. 5). The attenuator greatly reduces the generator signal level to verify the sensitivity of the measurement. The resulting signal, as measured on the analyzer, displays 4.56 µV average (5.06 µVRMS), which is correct (Fig. 6).
|
|
|
| Figure 6. | Designers can perform signal verification using the J2180A preamp with a 1X scope probe and a 20-dB gain correction factor to account for the 20-dB preamp gain. The measured 4.6-µV average signal is 40 dB above the noise floor of 46 nV and in excellent agreement with the 5-µVRMS signal level injected. The average to RMS conversion is 1.1, so 4.6 µV average × 1.1 =5.06 µVRMS. |
Having shown that the noise floor of the measurement is approximately 45 nV and verified that the setup correctly and accurately measured a 4.6-µV average signal, we can now use this setup to directly measure voltage regulator and reference noise.
Conclusions
We have demonstrated two simple methods for measuring power supply and reference noise with a Tektronix real-time spectrum analyzer. These methods provide significantly more information than the common oscilloscope, as they offer much greater sensitivity, as well as the particular frequencies that contribute the most noise. The addition of low-noise analog active filters and wideband preamplifiers, coming soon from Picotest, will bring further capabilities to this measurement technique, while further reducing the effective spurious response, particularly evident at 60 Hz.
