Power supply noise test

The distance between the GND of the probe and the two detection points of the signal is too large

    Figure 1: The quantization error of the oscilloscope DC. The oscilloscope has quantization errors. The ADC of the real-time oscilloscope is 8 bits, which converts the analog signal into 2 to the 8th power (that is, 256) quantization levels. When the displayed waveform only takes up a small part of the screen At this time, the quantization interval is increased, and the accuracy is reduced. For accurate measurement, you need to adjust the vertical scale of the oscilloscope (use variable gain when necessary), try to make the waveform fill the screen, and make full use of the ADC's vertical dynamic range. In Figure 1, the vertical scale of the blue waveform signal (C3) is a quarter of the red waveform (C2), and the rising edges of the two waveforms are magnified (F1=ZOOM(C2), F2=ZOOM(C3)), Then display the enlarged waveform with a long persistence. It can be seen that the waveform F1 in the upper right part has more steps (ie quantization levels), while the waveform F2 in the lower right part has fewer steps (ie fewer quantization levels). If you measure some vertical or horizontal parameters on the two waveforms C2 and C3, you can find that the standard deviation of the measured parameter statistics of the signal C2 that fills the screen is smaller than the latter. Explains the consistency and accuracy of the former measurement results.
    Usually to measure power supply noise, use active or passive probes to probe the power supply pin and ground pin of a certain chip, then set the oscilloscope to long persistence mode, and use two horizontal cursors to measure the peak-to-peak value of the power supply noise. One problem with this method is that the attenuation factor of a conventional passive probe or active probe is 10, and after connecting with an oscilloscope, the vertical scale is 20mV. When the DSP filtering algorithm is not used, the noise floor of the probe The peak-to-peak value is about 30mV. Take the 1.8V power supply voltage of DDR2 as an example, if it is calculated by 5%, the allowable power noise is 90mV, and the noise of the probe is close to 1/3 of the signal to be tested. Therefore, it is impossible to accurately test with a probe with a 10 times attenuation. Small voltages such as 1.8V/1.5V. When actually testing 1.8V noise, the vertical scale is usually between 5-10mV/div.
    In addition, the distance between the GND of the probe and the two detection points of the signal is also very important. When the two points are far apart, there will be a lot of
    Figure 2: The signal current loop on the probe EMI noise is radiated into the signal loop of the probe (as shown in Figure 2), the waveform observed by the oscilloscope includes other signal components, leading to erroneous test results. Therefore, it is necessary to minimize the distance between the probe signal and the detection point of the ground, and reduce the loop area.

    Figure 3: Schematic diagram of LeCroy PP066 probe For voltage testing of small power supplies, we recommend a passive transmission line probe with an attenuation factor of 1. When using this type of probe, the scale of the oscilloscope can reach 2mV/div, but its dynamic range is limited, and the adjustable range of offset is limited to +/-750mV. Therefore, when measuring common 1.5V and 1.8V power supplies, It needs DC-Block before inputting to the oscilloscope.
    Figure 3 shows the LeCroy PP066 probe. The distance between the probe's ground and the signal can be adjusted, and the probe's ground pin can be elastically contracted, which is very convenient to operate. Connect to the oscilloscope channel through a coaxial cable plus a direct-blocking module.
    You can also strip the coaxial cable, and directly solder the signal and ground of the cable to the power and ground of the power supply to be tested. In Figure 4, a section of the coaxial cable with the SMA connector was stripped and soldered to the 1.8V power supply of the DDR2 of the computer motherboard, and the power supply noise was measured.

    Figure 4: Measuring the 1.8V power noise of a computer motherboard DDR2

    After accurately measuring the waveform of the power supply noise, the peak-to-peak value of the noise can be calculated. If the power supply noise is too large, it is necessary to analyze which frequencies the noise comes from. At this time, it is necessary to perform FFT on the power supply noise waveform and convert it into a spectrum for analysis. The length of the signal time in the FFT determines the spectral resolution after the FFT. In the LeCroy oscilloscope, it supports the industry's 128M point FFT, which can accurately locate the frequency of the power supply noise (the spectral resolution is 40 times that of similar instruments the above).

    Figure 5: Measuring a 3.3V power supply noise

    Figure 5 shows the noise of the 3.3V power supply of an optical module. The frequency of the noise spectrum point is 311.6KHz. The same 312KHz periodic jitter was found in the jitter test of the 1.25Gbps optical signal output by this optical module. As can be seen in Figure 6, after decomposing the periodic jitter of the 1.25G serial signal (Pj breakdown menu), it is found that the periodic jitter of 312KHz is 63.7 picoseconds, and the jitter can also be clearly observed in the eye diagram. Through this explanation, power supply noise is likely to cause deterioration of the eye diagram and jitter of some high-speed signals. Spectrum analysis of power supply noise can effectively locate noise and guide the direction of debugging.
    When using an oscilloscope to measure power supply noise, in order to ensure measurement accuracy, it is necessary to select sufficient sampling rate and acquisition time.
    The recommended sampling rate is above 500MSa/s, so that the Necoste frequency is 250M, and the power supply noise below 250MHz can be measured. For the current popular board-level power integrity analysis, the 250M bandwidth is sufficient. Noise below this frequency can be filtered using ceramic capacitors, tightly coupled power supplies and ground planes on the PCB. However, the frequency above this can only be achieved by decoupling measures at the package and chip level.
    The longer the acquisition time of the waveform is, the smaller the spectral resolution (that is, delta f) after converting into a spectrum. Usually our switching power supply works above 10KHz. If the spectral resolution is to reach 100Hz, at least 10ms long waveforms need to be collected. At a sampling rate of 500MSa/s, the oscilloscope needs a storage depth of 500MSa/s * 10 ms = 5M pts.