The vigorous development of the Internet of things and intelligent terminal equipment has driven a series of applications such as intelligent driving, unmanned driving, unmanned aerial vehicles, intelligent robots and telemedicine. High performance chip is the brain of these applications, and its stability and reliability are directly related to the success or failure of the application. At the stage of product design and development, it is a major challenge for designers to quickly locate the accidental anomalies caused by chip and processor hardware or software.



How to deal with the measurement of ultra-low noise current in high-speed sampling, high dynamic and long-time is always a difficult problem. The extremely accidental abnormal current signal in MCU, FPGA and other devices has fatal risk to their devices. Solving this kind of accidental abnormal signal can timely make up for many deficiencies in noise, storage depth, waveform search and other aspects, and improve the reliability and competitiveness of devices.



Current characteristics of IOT products and challenges faced by traditional test instruments



Why should the primary goal be the analysis of current characteristics? The terminal devices for Internet of things, artificial intelligence, intelligent driving and intelligent medical applications are all facing the needs of long endurance and low power consumption. How to test and optimize the power consumption of devices is a big challenge. For such applications, especially smart wearable applications, the current of terminal devices has been very, very small, even to the micro ampere level. How to accurately measure the small current is a high challenge. Moreover, in the Internet of vehicles, MCU and ECU used in ADAS also need to measure and analyze the current waveform to gradually optimize the reliability of devices.





(key)


In this work cycle, we can expand the current waveform. The flow is about from the micro ampere level current of deep sleep to tens of milliamps or even hundreds of milliamps after wake-up, and then enter the sleep mode after receiving and sending. To optimize the overall power consumption, it is necessary to accurately measure the current characteristics under each operating state, and understand the normal operating state and potential problems. It is not difficult to see that the dynamic range of the current is very large, and the current signal frequency is high, and the pulse is very short, which is the challenge.



The traditional oscilloscope with differential voltage probe and sampling resistance can achieve good measurement, but it is still unable to accurately measure the clearly visible periodic acupuncture signal, which is mainly limited by the oscilloscope noise and ad bits. The digital multimeter (high-speed sampling) has no problem in terms of resolution, but its sampling rate is usually at the MS level, so it can not measure the spike signal (chopping pulse), and is insensitive to transient changes and transient anomalies.



How to quickly locate accidental signals by AI waveform analysis



The challenge mentioned above is that it is difficult to obtain accurate voltage and current waveforms. Another challenge is that the period is too short to capture waveforms long enough to capture accidental events. The third challenge is how to quickly identify accidental abnormal signals from a huge number of waveforms (millions of waveforms). The first challenge can be solved by reducing the bottom noise and improving the resolution; The second challenge is to complete long-time continuous waveform capture, which needs to solve the limitation of traditional measuring instruments by storage depth and measurement dead zone; The third challenge is to apply AI waveform recognition technology to complete rapid recognition from TB level waveform files.



To realize the intelligent waveform recognition in the last challenge, we must first solve the first two challenges. The first two challenges mainly put forward requirements for the key performance parameters of measuring instruments. First, the bottom noise (as low as 150pa/400nv) should be reduced as much as possible, and then the resolution (16bit) and dynamic range (100dB) should be improved. At the same time, sufficient signal bandwidth and sampling rate can be used to capture occasional transient signals.

(Intelligent waveform recognition process, key)


After capturing a large number of waveforms, directly save terabytes of data to an external mobile hard disk, and then start intelligent waveform recognition. AI waveform analysis function mainly includes powerful data and waveform comparison analysis, such as waveform playback, region amplification, trend analysis, FFT operation, etc.



After TB level waveform is triggered, trigger waveform similarity analysis and classification. AI recognizes and judges that the waveform abnormality is 90% waveform difference by default, and this accuracy can be adjusted to enter the depth screening. AI waveform analysis is directly embedded in the test equipment. This intelligent waveform recognition technology does not need to specify specific application scenarios, and data can be uploaded to other devices at high speed through external standard LAN and USB interfaces.



Summary



The premise of AI waveform recognition technology is a large number of waveform storage and waveform distortion free measurement, which requires the test equipment to have a low enough bottom noise, high enough resolution, dynamic range, signal bandwidth and sampling rate to support the subsequent AI waveform analysis. The integration of AI technology greatly improves the efficiency of the test process, and the guarantee of accuracy also avoids possible omissions in the traditional measurement analysis, which is very helpful to improve the reliability of devices.

Leave a Reply

Your email address will not be published. Required fields are marked *