X5-RX timing and channel-to-channel registration

[mrodby]

We are using an X5-RX in a system where registration between channels needs to be much smaller than a sample time in order for our analysis algorithms to work properly. That is, given identical signals, the results from any two channels should be substantially identical. Unfortunately, we are finding that our X5-RX does not meet this requirement.

In our test setup, we feed a single sine wave into a splitter, and then route the two splitter outputs into two channels of the X5-RX, through identical cables that are within a small fraction of an inch of the same length. All of our tests currently use a sample rate of 130 MSPS, so the sample time is just under 8 ns. We then record about 1 ms of samples from both channels interleaved, write the results to a file, and read the file in to MATLAB. Since our X5-RX is not calibrated, we subtract the average sample value from all samples (considering each channel separately), and then compare the zero crossings of the two channels. Though the amplitudes of the two channels are different, once the average sample value has been subtracted, the zero crossings should occur at the same time for both channels.

The largest difference we have seen is between channels 0 and 1, with a 250 kHz sine wave, where the difference was about 100 ns, or about 12 sample times. The other differences have been on the order of 1 to 1.5 sample times. At 10 MHz, the difference between channels 0 and 1 was about 1 sample time, as was the difference between channels 2 and 3. At 20 MHz, 40 MHz, and 50 MHz the difference between channels 0 and 1 was about 1 to 1.5 sample times; the difference between channels 2 and 3 was essentially zero at these same frequencies. At 50 MHz, the difference between channels 0 and 1 is over 180 degrees, which makes this board essentially unusable for our application.

All of these are repeatable, with only tiny changes from one test to another at the same frequency.

How should I interpret these results? Are these results reasonable to expect with the X5-RX, or could we have a defective board?

[jhenderson]

We are preparing an experiment to investigate this matter. We expect to obtain some results today and will respond with our diagnosis asap.

The phase and amplitude of input signals less than 1 MHz passing through the high-speed A/D chain will deteriorate due to the characteristics of the input transformer. So, the minimum frequency of input signals digitized by the high-speed A/D should be at least 1 MHz. Use of a square instead of a sine wave should simplify the phase measurement, since it a square wave would not be affected by small bias errors on individual channels.

Determining the phase of a 50 MHz input signal asynchronous to the sample clock must be calculated carefully, since the number of points per waveform cycle is small. I suspect this is the cause of the phase mismatch observed with this test case.

[mrodby]

It has been 2 weeks since you expected to have some results. Were there unexpected results that you are still investigating, or did some other priority interfere with the testing?

[jhenderson]

I thought that this issue was closed. I posted the results of the testing at http://www.innovative-dsp.com/ftp/X5-RX/Phase%20Alignment.zip a couple of weeks ago. Unfortunately, it appears that you were never notified properly.

The results in the zip referenced above show that we clearly maintain phase alignment. As the number of samples per wave decreases, it becomes difficult to determine the absolute phase. Since by design, all A/Ds in the system receive a common clock and tests at lower frequencies prove phase preservation, we concluded that this is not an issue.

[mrodby]

Since I am using Linux, and the files you posted require using BinView, I tried installing BinView from the web. It appeared to install, but when I ran it, an error message appeared:

This application has failed to start because IPPS-5.3.DLL was not found. Re-installing the application may fix this problem.

I could not find IPPS-5.3.DLL anywhere on my system, nor in any of the files in the BinView installation. I have tried re-downloading it and reinstalling it, and even tried downloading and installing on a different computer, all with the same result.

If this is a problem with the BinView installation files, please let me know when it is fixed. If it is something I am doing wrong, please let me know that too.

[jhenderson]

At the current time, Binview is a Windows-only application. Install it under Windows and also install the MalibuRed package from http://www.innovative-dsp.com/cgi-bin/dl6.cgi?product=MalibuRed

[mrodby]

Thank you for posting the data, and for your help getting BinView to work.

As with the data I obtained from my X5-RX, the data you posted shows a consistent time shift between channels, with a different time shift at each frequency. Though your time shifts are smaller than mine, they are still larger than we can tolerate.

The most striking example of this is in Ch2&Ch3_10MHz.bin. Open this file in BinView, view the time domain signal, maximize the window, and set the span to a small number (e.g. 100) so that you can see each sample clearly. When the sine waves cross zero, channel 2 (shown in BinView as channel 0) is almost 2/3 of a sample time ahead of channel 3 (shown as channel 1), which is equivalent to about 3 ns. This relationship is remarkably consistent through the entire file. The difference in CH2&Ch3_1MHz.bin is even larger: between 1 and 1.2 sample times, or about 5-6 ns. Again, the relationship is remarkably consistent throughout the entire file. Note that because the gains of the two channels are different, the only time you should compare the timing is near a zero crossing. Otherwise the time offset is distorted by the amplitude difference. But even when the gains of the two channels are significantly different from each other, zero crossings should happen simultaneously.

In an earlier comment, you mentioned that you thought the effect I was seeing was probably caused by the small number of samples per signal cycle. If that were the case, the shift between channels would change as the sample points occur at different points along the sine waves. With such a consistent time difference (consistent in direction and magnitude), I can’t see any conclusion other than that it represents a real time shift between channels, or a phase shift in the analog circuitry that differs between channels and at different frequencies.

If the time shift were consistent from one frequency to another, we might be able to calibrate it out, though that might take more real-time computation power than we have available. But since the time shift varies depending on the frequency, that takes the problem beyond what we can correct, so it makes this board almost useless in our application. We need the time shift to be significantly less than 1 ns in order for the phase of high frequency signals (~50 MHz) in quadrature to be accurate.

If you have a different interpretation of the data you posted, please elaborate.

[mrodby]

[jhenderson]

We are willing to exchange the X5-RX with a DC-coupled X5-210M, which does not use a separate A/D circuit to provide DC- information. The DC coupled 210M would have excellent phase matching across channels and is largely software compatible with the RX. Call the factory to discuss, if interested.

[mrodby]

Thanks for the offer, but we need all of the resolution we can get. The drop from 16 to 14 bits would reduce our sensitivity too much.

Any chance we will see a 16 bit DC coupled ADC that will do at least 130 MSPS continuously from Innovative any time soon?

[jhenderson]

Yes. The new X6-RX module is in PCB layout now. A preliminary data sheet is available at http://www.innovative-dsp.com/products.php?product=X6-RX. We expect to begin shipping at the end of July.

Correction: X6-RX is AC-coupled. Please ignore.

[CodyPylant]

[jhenderson]

We do not have one planned, but feel free to call and discuss. If a business case exists for this capability, we can certainly spin one of the existing designs to meet your requirement.