开云体育

I have very little knowledge of DSP, so would appreciate if anyone here can help out.

One of our club members M0KGW suggested that the RBW of the tinySA or homebrew simpleSA could be improved using convolution.

The idea is if you know the response of the SI4432 FIR filters, (we can measure this) then by sampling at smaller frequency step increments and using convolution then it should be possible to recover some of the information to get improved resolution bandwidth (RBW).? Improved RBW is desirable for measuring intermodulation of transmitters.

Consider say two signals at close spacing, say 1kHz apart.? The SI4432 has a minimum RBW of 2.6kHz so by stepping at say 2kHz intervals it is not possible to pull out these two signals.? If a smaller frequency step is used, then as the signals come in and out of the filter passband the RSSI value will change - the RSSI at each frequency step will be the sum of the filter response for each signal.? Given the frequency response of the filters it should be possible to extract the two signals using convolution in the frequency domain.

45 years ago I might have been able to work out the maths for this, sadly no longer.? Also sad to say I had not even heard of convolution before M0KGW mentioned it.

Is this idea practical and if so how to go about coding it?? I would certainly like to give it a go and see how it works in practice.

73

Dave M0WID

It's actually deconvolution ( ?) and it can work as the tinySA RBW filters are well defined and are fairly steep.
The required memory may be too high and the frequency resolution is only 260Hz allowing maximum 10 substeps.
The required scan time with 10 substeps is at least a factor 10 higher.
I may look into this but do not have too much hope.
--
HBTE Files section:?/g/HBTE/files
Erik, PD0EK

On Tue, May 11, 2021 at 08:45 AM, m0wid wrote:

sampling at smaller frequency step increments and using convolution

Interesting. This sounds like a one dimensional version of what Photoshop does to photos to sharpen up somewhat blurry photos in two dimensions.?

Not introducing artifacts might be the biggest challenge. But, you did say the response of the SI4432 is important to measure as part of the process.?

Tom, wb6b

I've implemented and tested deconvolution.
Step 1: Measure the response to a single signal and use that as the deconvolution filter
Step 2: Measure the convoluted response to multiple signals
Step 3: deconvolute using fft
The outcome was not usable due to the noise in the input.
--
HBTE Files section:?/g/HBTE/files
Erik, PD0EK