That sounds like an ambitious project. Off the top of my head, it sounds doable. You’d have to generate a CSV file with your two tones, measure the amplitude and phase of the 3rd order responses (in a highly linear measurement system), and then subtract the 3rd order responses from your CSV file (or something similar). I would expect that you could drop your 3rd order intermod products by 20 dB at least, but you’d have to try it to know. And it would vary with temperature and frequency, so you would probably want to do it right before using it or after a 30 minute warmup. Assuming you want tones spaced 2 MHz (+/- 1 MHz) you’d want to choose a DAC clock that is a multiple of 6 * tone spacing. Something like 96 MHz would probably work nicely.
If you can’t get phase info, you could use the amplitude of the IM products and hunt for the right phase that exactly cancels the undesired IM products.

I see you’ve put some thought into this.
1) The RBW calculation you gave is actually the bin size calculation. RBW is bin size * FFT window bandwidth. Typically, a window bandwidth of about 2 is required or spectrum analysis, for a 30k RBW.
2) The bin size will be as the power-of-2 math determines. Rarely a nice number.
3) We will have to drop to 16 bit amplitude data for these sweeps, typically 0.01 dB resolution.
4) The 120 us includes tune time of about 20 us.

Sure. The SM200A has the ability to perform FFTs on about 500 million data points per second, but the maximum FFT size in our design is 16384. The dual channel 500 MSPS ADC is run through a FIR correction filter, a digital single sideband select filter, and then converted to 250 MSPS I/Q data (160 MHz). This can then be processed with the on-board FFT engine, or decimated further to 50 MSPS (40 MHz) and streamed to the PC, where much larger FFTs are possible.
Yes, the limitation for narrower RBWs is the USB3 speed. For FFTs on the FPGA, we impose a 120 microsecond minimum dwell time at each LO, which becomes the limiting factor, but for a 30k RBW, USB3 speed is nearly a limiting factor even for basic amplitude data.

You also mentioned sweep speed at 1 kHz RBW. While we should have 1 THz/sec at 30 kHz RBW, once you cross into the 40 MHz patches, this slows down. You could expect about a 2-4 ms acquisition time per patch for a 1 kHz RBW. Assuming your CPU was decent, this would give you 40 MHz / 4 ms, or about 10 GHz/sec (maybe ~2 seconds for a 20 GHz sweep)

Mikes,
We have been using the 2/3 (IP3-DANL@1Hz RBW) for dynamic range.
You mentioned a -130 dBm signal with a -20 dBm signal. The noise floor for a -20 dBm signal and reference level would be about -157 dBm/Hz, so I’d say you’d have to have a 100 Hz RBW to clearly see the -130 dBm signal, and you’d have to ensure phase noise wasn’t a limiting factor. At 1 GHz, internal phase noise would not block a -110 dBc signal if you were > 300 Hz from a clean -20 dBm CW blocker. The phase noise should be better than -133 dBc/Hz (>= 10k offset) at 1 GHz, but this goes down 20 dB / decade as you increase frequency.
Also, for 160 MHz patches, you are limited to >= 30 kHz RBW. For lower RBW, 40 MHz patches are used.

David,
It sounds like whenever it tries to download the calibration table, it hangs. We may need to take a look at it here. You can contact support at signalhound dot com for an RMA. I apologize for the inconvenience. If you purchased it less than 2 years ago, it should be covered under warranty.

Thank you for the feedback and tips!

The SA44B / TG44A with coupler can be used to measure magnitude S11 and magnitude S21. It cannot provide the phase of S11 or S21. This is the critical difference between a VNA and SNA.

A two port VNA can also provide S22 and S12 without swapping your device around.

The best way to mute the output is to set the frequency to 0 Hz, amplitude to -30 dBm.

Eduk,
We have observed, when the preamp is used, at 0 dB of attenuation, it can take a long time for the DC blocking cap to reach steady state (up to 30 seconds). Typically, setting your reference level to -30 dBm briefly (preamp on, attenuator 5-10 dB), before proceeding to maximum sensitivity, allows quick settling.

On a fraction of units, the preamp’s DC blocking cap never reaches steady state without a better DC path to ground. We typically have to add a 2.2k resistor to help bleed this off. You can contact us for a warranty repair if your SA44B is exhibiting this behavior. I apologize for the inconvenience.

We will be updating the design to add this resistor in the near future.

The SA44B definitely has some shortcomings when it comes to EMC testing. It relies on a software image rejection algorithm to get rid of both image responses and its own spurious responses. This algorithm tends to work better on sparse spectrum with narrow RBWs, where the chance of a spur and an image overlapping is low. When there are many signals closely spaced, especially at low frequency where there may be higher frequencies present mixing with harmonics of the LO, the chance of spurious signals overlapping on the primary and image sweep are high.

This is one of the reasons we strongly recommend the BB60C for EMC work. With hardware image rejection and specified spurious performance, you get reliable EMC sweeps.

We aren’t going to IWCE, but we will be at IMS in June.

If you’re building CW / FM waveforms in excel, you can generate the instantaneous phase data, and then use the sin and cos functions to convert to I/Q data.

The .csv file is generally for non-standard waveforms. Our API can be used to control the VSG without the GUI.
The CSV file can be used if you have less than 2048 I/Q samples you wish to send, either as a repeating burst, or a continuous loop.

Jaguirre,
From a hardware standpoint, the SA44 does not have a preamplifier, so the noise figure is much higher. It also has more spurious and residual signals than the SA44B. But it should run in our Spike software. There are additional RBW / VBW limitations for the SA44 as well, for spans larger than 200 kHz.

Eduk,
The SA44B works by capturing 200 kHz patches of spectrum and processing them. There seems to be some 200 kHz periodicity. I noticed that the peak-to-peak periodicity improves for the stronger signal, so it looks like what you are seeing is the IF filter shape on the SA44B’s noise level.

Because your RBW is 100 kHz, the software has a hard time correcting for IF flatness. If you reduce your resolution bandwidth to 30 kHz, the software will have an easier time correcting the IF flatness, and this ripple should be reduced.

The SA44B requires 480 MBPS USB (high speed USB 2.0). It also requires about 400 mA of current from the port. If the isolated hub cannot provide the data rate and required power, it would not work. Otherwise it should be fine.

The “uncalibrated” indicator means that a valid “store thru” has not been performed, or a setting was changed after it was performed, invalidating it.

If you set the desired center frequency, span, and reference level, select the appropriate scalar network analysis settings, and then store the thru, this indicator should remain off until you change a setting.

Hopefully this helps.

Jason,
The biggest issue is that the ARM architectures we have tested cannot stream I/Q data at the full data rate, so there are RBW / VBW limitations. Additionally, FFTs take much longer, so on the ARM architecture we do the minimum number to satisfy the settings.
For better performance, something like an Intel Atom processor gives you the full SA44B performance at lower power consumption.

I’m sorry you had a hard time with the drivers. Lately, many customers have had to un-check the “Load VCP driver” check box, but that is usually the extent of the driver problems.

If this happens again, there is a second possibility. If the SA124A has trouble booting or rebooting, there is a small FTDI settings patch I can send you that might help.

[Author]

Posts