Use oscilloscope frequency domain method to analyze power supply noise

Application of oscilloscope frequency domain analysis in power supply debugging
This article talks about the power supply noise measurement problem that has been concerned for so many years. It has practical experience summaries, actual test evidence, and a combination of simulation analysis.
In the analysis process of power supply noise, the more classic method is to use an oscilloscope to observe the power supply noise waveform and measure its amplitude to determine the power supply noise. However, as the voltage of digital devices gradually decreases and the current gradually increases, power supply design becomes more difficult, and more effective testing methods are needed to evaluate power supply noise. This article is one of using the frequency domain method to analyze the power supply noise. When observing the time domain waveform can not locate the fault, use the FFT (Fast Fourier Transform) method to perform time-frequency conversion, and convert the time-domain power supply noise waveform to the frequency domain for analysis. During circuit debugging, the signal characteristics can be viewed from the two perspectives of time domain and frequency domain, which can effectively speed up the debugging process.
During the board debugging process, it was found that the power supply noise of a network reached 80mv, which has exceeded the device requirements. In order to ensure the stable operation of the device, the power supply noise must be reduced.

Review the principle of power supply noise suppression before debugging the fault. As shown in the figure below, different frequency bands in the power distribution network have different components to suppress noise. Decoupling components include power adjustment modules (VRM), decoupling capacitors, PCB power ground plane pairs, device packages and chips. The VRM includes the power chip and the peripheral output capacitor, which acts approximately in the DC to low frequency range (about 100K). Its equivalent model is a two-element model composed of a resistor and an inductance. The decoupling capacitors are used in conjunction with capacitors of multiple orders of magnitude to fully cover the mid-range frequency range (about 10K to 100M). Due to the presence of wiring inductance and package inductance, even a large number of decoupling capacitors are difficult to function at higher frequencies. The PCB power ground plane forms a plate capacitor, which also has a decoupling effect, which is about tens of megabytes. Chip packaging and chips are responsible for high frequency bands (above 100M). Current high-end devices generally add decoupling capacitors to the package. At this time, the decoupling range on the PCB can be reduced to tens of megabytes or even several megabytes. Therefore, when the current load remains unchanged, we only need to determine which frequency band the voltage noise appears in, and then optimize the decoupling component corresponding to this frequency band. The two decoupling elements will cooperate in the adjacent frequency bands of the two decoupling elements, so the decoupling elements of adjacent frequency bands should also be taken into consideration when analyzing the critical point of the decoupling element.

According to the traditional power debugging experience, first add some decoupling capacitors to the network to increase the impedance margin of the power network, and ensure that the power network impedance in the mid-frequency band can meet the needs of the application scenario. As a result, the ripple is only reduced by a few mV, and the improvement is minimal. There are several possibilities for this result: 1. The noise is at a low frequency and is not within the range where these decoupling capacitors work; 2. The increase in capacitance affects the loop characteristics of the power regulator VRM, and the impedance reduction caused by the capacitance is related to the VRM. The deterioration was offset. With this question in mind, we consider using the frequency domain analysis function of the oscilloscope to view the spectral characteristics of power supply noise and locate the source of the problem.
The frequency domain analysis function of the oscilloscope is realized by Fourier transform. The essence of Fourier transform is that any sequence in the time domain can be expressed as an infinite superposition of sine wave signals of different frequencies. We analyze the frequency, amplitude and phase information of these sine waves, which is the analysis method of switching the time domain signal to the frequency domain. The sequence sampled by the digital oscilloscope is a discrete sequence, so the fast Fourier transform (FFT) is commonly used in our analysis. The FFT algorithm is optimized from the Discrete Fourier Transform (DFT) algorithm. The amount of calculation is reduced by several orders of magnitude. The more points that need to be calculated, the greater the savings in calculation.
The noise waveform captured by the oscilloscope is transformed by FFT, and there are several key points that need attention.
1. According to Nyquist's sampling law, the spectral spread (Span) after transformation corresponds to 1/2 of the sampling rate of the original signal. If the sampling rate of the original signal is 1GS/s, the spectral spread after FFT is mostly 500MHz ；
2. The frequency resolution after transformation (RBW Resolution Bandwidth) corresponds to the reciprocal of the sampling time. If the sampling time is 10mS, the corresponding frequency resolution is 100Hz;
3. Spectrum leakage, that is, the interference between the various spectral lines in the signal spectrum, and the lower-energy spectral lines are easily overwhelmed by the leakage of adjacent high-energy spectral lines. To avoid spectrum leakage, try to synchronize the acquisition rate with the signal frequency, extend the acquisition time and use an appropriate window function.
The power supply noise measurement does not require a high sampling rate, so a long time base can be set. This also means that the collected signal time can be long enough to cover the entire effective signal time span. There is no need to add a window at this time. function. Adjust the above settings to get a more accurate FFT transformation curve, and then use the zoom function to view the frequency points of interest. As shown in the figure below, the main energy of power supply noise is concentrated around 11.3KHz, and this frequency is used as the fundamental frequency to resonate. Based on this, it can be inferred that the impedance of the PDN network at 11.3KHz cannot meet the requirements, and the impedance of the capacitor at this frequency point is also relatively high, which can not reduce the impedance, so the previous increase of the capacitor cannot reduce the power supply noise.

Generally speaking, 11.3KHz should be the jurisdiction of VRM. The larger noise here indicates that the VRM circuit design cannot meet the requirements. The performance of VRM is analyzed here. There are many methods of VRM analysis, and the method of simulating the Bode plot of its feedback loop is mainly used here. The Bode diagram mainly observes several key information: 1. Crossover frequency, the frequency point where the gain curve crosses the 0dB line; 2. Phase margin, the phase value corresponding to the phase curve at the crossover frequency; 3. Gain margin, the phase is at- The gain value corresponding to 360°. Here we mainly focus on the two indicators of crossover frequency and phase margin. From the loop Bode diagram of the VRM (as shown in the figure a below), it can be seen that the crossover frequency of the VRM is around 8KHz, and the phase margin is 37 degrees. There are two problems here: First, the phase margin of the VRM generally needs to be greater than 45 degrees to ensure the stable operation of the loop. Here the phase margin is slightly smaller, and the phase margin needs to be increased; secondly, the crossover frequency is too low, and the VRM near the crossover frequency The adjustment effect is gradually reduced, and the bulk capacitor at this frequency point does not work, so there will be a higher impedance near 8KHz, and the noise suppression effect at this frequency point is poor. The following figure (b) is the Bode plot after optimizing the VRM loop, adjust the phase margin to 50 degrees, and push the crossover frequency to about 46KHz.

(A) Original waveform

(B) Waveform after optimization

Verify the ripple for the optimized VRM, and it can be seen that the ripple is significantly reduced to 33mv, which can meet the requirements of the device.

The above is the process of using the oscilloscope FFT function to quickly locate the power supply problem. From this example, you can see that the oscilloscope's frequency domain analysis function can play a great role in circuit debugging. The FFT function of the oscilloscope with long storage depth can easily analyze low-frequency and long-period signals. This advantage is more prominent in digital circuit debugging.