Contact an Expert

Can you provide a brief overview of "Measurement Process" in the Agilent Dynamic Signal Analyzers (DSA) such as the 35670a and 35665A?

This paper attempts to briefly explain how Hewlett-Packard/Agilent DSA analyzers operate. This is directly applicable to the 35665A and 35670A. By experimentation, it appears that the 3562A and 3563A operate the same way. The normal measurement process is described in steps 1 through 9. Step 1 describes calibration. Step 10 describes possible results after the Inverse FFT (IFFT) of a linear spectrum measurement result.

  1. Regularly, automatically, the analyzer performs a calibration operation on its front-end circuitry. This verifies proper operation and provides frequency response data that will be used with each FFT result to correct the measurement results. Additionally, the instrument firmware contains detailed information about the digital signal processing algorithms and calibration data to correct for its effects. Calibration is performed at set intervals after power is applied, primarily to correct for temperature drift effects on the analog front end. When input range is changed or analysis span is changed, new calibration curves are calculated for use in the correction of the frequency domain FFT result.
  2. Wait for the trigger point. This trigger point is always referenced to the beginning of a time record. Since the actual trigger point will usually fall between samples in the time record, a 10 MHz counter is used to find the actual trigger point in time relative to the leading time sample. The trigger counter value will be used later in the process as part of phase correction. The counter value is not available to the user.
  3. Collect a time record. For each channel the DSA uses a single fixed analog anti-alias filter and its digitizer operates at a fixed sample rate. Reducing the sample rate to provide a reduced measurement span is accomplished using digital filtering and decimation. There will be corrections made later to accommodate the magnitude and phase effects of the input circuitry and digital filtering.
  4. Apply the time domain window function to the time record. For the Hanning or flattop windows, the beginning and end of the time record are reduced to zero. If the Uniform window is selected, no change is made to the data.
  5. Move time record data. Only if a Hanning or Flattop window was applied, move the first half of the windowed time record to the end of time record, and move the second half of the time record to the beginning. This puts the zeroed, non-contiguous in time, data in the middle of the time record. This data switching is done because the FFT assumes that the data in the time domain is continuous; it assumes that the first sample and the last sample are contiguous. This is not true in the real world. The Hanning or Flattop window zeroes the non-contiguous parts of the wave, the original beginning and end of the time record. Then, in order to get a significant signal for phase analysis by the FFT, we move the center of the time record to the beginning, which is what the FFT uses as the phase reference.
  6. Perform the FFT operation. This produces real and imaginary frequency domain data, which can eventually be displayed as magnitude and phase. The phase reference for this data is the first time domain sample in the data provided to the FFT. In all cases, the phase reference is a cosine wave. This means that if the uniform window is selected, the phase reference is the first sample of the record collected in step 2. More importantly, if the Hanning or Flattop is selected, the phase reference is at the middle of the time record.
  7. Apply the phase correction from the trigger counter. This is caused by the time delay between a sample point and the actual trigger point. This correction can be the result of up to ± (time between samples) in the time record.
  8. Apply the dynamic correction for the analog front end and the digital filtering. This is a magnitude and phase correction, dependent upon the instrument setup, particularly the input range and measurement span.
  9. Display corrected linear spectra, power spectra, real, imaginary, etc., in the frequency domain. This is the normal operation of the analyzer.
  10. If a Hanning or Flattop window was applied in step 4, then an IFFT of a linear spectrum from step 9 will produce a time record that is not useful (the window function cannot be removed accurately). If the uniform window was used in step 4, then the IFFT will produce a useable time record, corrected for the trigger point and analog and digital processing. The 35670A allows the user to turn off the calibration calculation part of the measurement algorithm with the following keystrokes: [System Utility] {More} {Service Test} {Spcl Test Modes} {More Spcl Modes} {Cal Const OFF}. If Cal Const is OFF, and the Uniform window is selected, the frequency domain result (Step 9) is no longer calibrated, and an IFFT of the linear spectrum produces exactly the same data as the time record from step 3.