Forum Replies Created
- Posts
Matthew,
I have never seen that before, but if you can write a little code, parse the wav file, probably window the data, do an FFT, threshold detect (1 for above threshold, 0 for below threshold), invert the threshold array, inverse FFT, then scale. I am not aware of an application for this, but if you have one, something like this would be a good way forward.Geev,
The I/Q pattern memory is pretty short (2k samples), but yes, a short Gaussian waveform is possible. You can also look at 1001 tones with random phase relationship, which will have many of the same properties as bandwidth-limited Gaussian noise.For a better Gaussian white noise generator, consider the VSG60A.
Kefei,
There are a few ways to do this. You could record the sweeps in Spike (you may need an SM200 in real-time mode, span >160 MHz, for 1 ms sweeps).
For more control, you could simply record 1 second of I/Q and do your own windowed FFTs. With the SM200B, this would give you up to 2 seconds at 160 MHz span.
A third option is the waterfall spectrum plot in zero span mode. In this mode, the biggest limitation is probably the 1 second requirement. This would set a pretty low I/Q sample rate, since you are limited to a few million I/Q data points.Volker,
Yes, this is something we have discussed. It is not something we have scheduled at this point, but it is definitely something we hope to add in the future. It would probably not be one of our existing tracking generators, but something geared towards the frequency range and performance of the SM series.Gary,
There are a few things going on here. It sounds like you are aware there is no baseband reconstruction filter on the VSG25A. But we also apply an inverse sinc correction to the in-band data to flatten it.
PN9 is a PRBS9 sequence (linear feedback shift register) that basically repeats every 511 symbols. We upsample each symbol by a factor of 4 to produce 2044 I/Q samples, which get transmitted in a loop.
Even with no raised cosine filter, the upsample process will result in gaps in the sin(x)/x curve.
Hopefully this helps.If it helps there are schematics online. Typically a 1 Mohm input impedance to a non-inverting unity gain op amp, with a 50 ohm series resistor (or resistor to bring it to 50 ohms) on the output. You may want to add a ceramic [0.1 – 10] uF DC blocking cap, depending on your low frequency coverage needs.
We have a probe that works from about 100 kHz to 4 GHz, with about a 500 ohm AC load to your circuit (DC blocked). We are not set up for a FET probe, which would be what you would need for a high impedance load to your circuit but a 50 ohm output impedance. This is more of an oscilloscope function.
Kefei,
To convert I/Q data to spectrum, you will need to perform overlapping windowed FFTs. Typically what this looks like, if you are doing a 1024-point complex FFT, is every N samples (for your application this would be every 1 ms), you will take the last 1024 I/Q samples, window them (multiply by a window function, like a flat top or hamming window), and then FFT them to generate a spectrum.
We have not interfaced SDR sharp. One approach would be for your friend to develop a plugin, to connect the BB60C to SDR# directly. Another would be to use our API to save I/Q data as a WAV file and try to play that in SDR#.
Basically, queued fast sweeps just go. Unfortunately there is a little bit of uncertainty each time the LO changes. We check to see if it has settled, and add a few microseconds delay if it has not, so even if they started in sync it would not last.
We could add this to our list of feature requests. Most likely this would be a feature we would add to a future product (multi-channel receiver) due to the complexities of synchronizing two standalone devices.
If you could scan your spectrum more slowly, 40 MHz at a time, you could use I/Q streaming and a shared external trigger to get them sync’d within a handful of nanoseconds.
Keysight has a good IP3 guide: https://www.keysight.com/upload/cmc_upload/All/ENA_IMD_Measurement_Summary.pdf
That model should work just fine. A 10 MHz square wave will provide slightly better performance, but it looks like the output is +10 dBm or higher, which is good.
For what it’s worth, I get 375 ms on my PC for the same sweep. Unfortunately, sweep time, especially on narrow spans, is not deterministic. There are settings applied, and then acknowledged, several times per sweep. So the timing will vary based on your USB read/write latency, which can vary for a lot of reasons. Spans of several MHz will generally be more consistent since the sweep is controlled more in firmware than software.
Do you mind if I ask what the center frequency, span, RBW and VBW settings are? What is the model of the i7 processors? Is spur reject on for both cases?
It is possible for a sweep to take twice as long if the PC is significantly slower, and smaller variations are fairly common, but sweeping twice as fast on a similar PC seems unusual.
Around 2016 we started adding a resistor to provide a weak return path to ground because some units would develop this condition you described over time. I think the problem stems from the fact that there’s no return path to ground when the attenuator is 0 dB, and a tiny leakage current from the attenuator can essentially blind the RF switch until a ground path exists.
If you can set the attenuator to 5 dB, or add a 1 dB external attenuator, this problem should clear up. Otherwise, you can get an RMA, send it in, and we can add the resistor.
Cristin,
There are several factors that go in to why these measurements are different. These include the software image rejection algorithm, the noise floor / sensitivity of the SA44B in the selected mode, and the effective video bandwidth and averaging method based on detector settings.
If your goal is noise spectral density measurements, the preferred instrument is the BB60C or SM200A. This is because the SA44B does not have hardware-based image rejection, so there is noise present from both sidebands (even if the signal from one sideband is rejected in software).
That being said, real-time mode, power average is a good way of measuring the noise energy (sum of energy at desired frequency and image frequency–subtract 3 dB typically for noise at desired frequency only).
For a lower noise floor, using sweep mode (image rejection ON), power, average, VBW <= 1/10 RBW, will use a more sensitive detector circuit and provide a lower noise floor than real-time mode.Miran,
One option for powering down the SM200B is to shut down the PC controlling it. Most PCs allow you to select whether or not the USB is powered when it is shut down. If the USB 5V is turned off, the SM200B will turn off completely (I believe it is something like a milliwatt of power draw).
If a mechanical switch is what you want, any switch that is rated for at least 20V, 5A should be sufficient.
If an approximate conversion loss (+/- 3 dB) is acceptable, you can sweep the output range with the SA44B, and then step across the input range with the TG manual controls, using MAX HOLD on a trace.
For better accuracy, an RF power meter could be used to measure the actual TG output power at each frequency and you could subtract this from the SA44B’s reading.
The GPIO is driven with two, SN74AVC4T774 chips. 10k pulldown, 100 ohm series resistor. The 100 ohm resistor will limit your switching rate as you add capacitive loading, and you might have to worry about ground bounce, etc. with long cables. But this should give you enough to model it if you have concerns.
A fan-out buffer board could always be the fall-back position. I’d use something like a Max 10, so that you could get creative if you needed to drive all 60+ devices independently (e.g. one nibble is data, the other nibble is address for 64 independent outputs)
It looks like some strong VSWR effects. I would throw a 6 or 10 dB attenuator on the SA and TG and repeat the test. If this does not solve the problem, I would try different cables.