I am given a recording of an ionosonde sounding. The signal $S$ is represented as a series of I/Q pairs at a sample rate $F_{sr} = 10\,\text{MHz}$. I am told that the center frequency of the recording is $F_{center} = 7\,\text{MHz}$.

I have a $T=200\,\text s$ recording as $I=[I[0],..,I[T\times F_{sr}-1]]$, $Q=[Q[0],..,Q[T\times F_{sr}-1]]$.

The actual signal $R[t]$ being recorded can vary from $2\,\text{MHz}$ to $12\,\text{MHz}$, that is from ${F_{sr}\over 2} -F_{center}$ to ${F_{sr}\over 2}+F_{center}$.

Let $S=I+Q j$ be the complex representation of the recording.

How do I recover the original signal $R[t]$ at a sample rate of $F_{up}=24\,\text{MHz}$, which is high enough to represent a $12\,\text{MHz}$ signal directly, as a function $R(S,F_{center},F_{sr})$?

That is, $F_{up}=2F_{sr}+ 2 F_{center}$, so $R=R[0],...,R[T\times F_{up}-1]$.

  • $\begingroup$ I'm not quite sure what you mean with "$R$ can vary from 2 to 12 MHz", because that's actually exactly the bandwidth represented by complex sampling of a baseband signal representing the spectrum around 7 MHz at a rate of 10 MS/s. What is "the original $R$"? To me, the original signal is your $S[n] = I[n] + jQ[n]$. What'd be "more" original than that? $\endgroup$ – Marcus Müller Jun 18 '19 at 7:12
  • $\begingroup$ The "original R" is the analog signal captured from the antenna of a Lowell Digisonde 4D (digisonde.com/pdfs/Digisonde4DManual_LDI-web1-2-6.pdf) and then digitized as IQ pairs into an 8GB bin file for the IARPA PINS Challenge (iarpa.gov/challenges/pins.html). I want to look at the spectrogram of the whole signal, as in Figure 2 of this writeup: topcoder.com/challenges/30088355. I thought it would be easier to make Figure 2 if I had the real signal in the original upsampled rate. When I make a spectrogram of the IQ signal it doesn't look like Figure 2. $\endgroup$ – Lars Ericson Jun 18 '19 at 13:27
  • $\begingroup$ But the analog signal that gets digitized is the I and the Q signal – the RF bandpass signal doesn't exist anywhere, digitally, and contains no additional info compared to the baseband signal – only a relabeled frequency axis. $\endgroup$ – Marcus Müller Jun 19 '19 at 6:19
  • $\begingroup$ I was going through this exercise because when I do a spectrogram of the recording I just get a very bandy picture, where Figure 2 has clear tracks for the chirps. I found an article on Chirp Reception and Interpretation by a Dutch ham radio enthusiast (websdr.ewi.utwente.nl:8901/chirps/article) which, under the heading "Supresssing non-chirping signals", shows a bandy picture and then gives a good trick: Null out all the constant frequency high-power bands, and the chirps will remain. This is under the heading of "SDR chirpfilters", which I have to look up. $\endgroup$ – Lars Ericson Jun 19 '19 at 11:09
  • $\begingroup$ Also by the way the IARPA PINS challenge is still open for another week or so,. This question is actually worth $25,000. I won't be sad if you jump in and grab that easy money ahead of me. $\endgroup$ – Lars Ericson Jun 19 '19 at 11:11

I gather you are trying to reconstruction a real (non-imaginary) version of your I & Q signal. The following scheme should work:

  1. Up sample your IQ from 10 MHz to 24 MHz. This involves zero padding and applying a low-pass filter to remove the extra images that appear due to the zero padding. See below for more details.
  2. Mix the signal with $\exp(j2\pi7\times 10^6t)$. This shifts the signal to the 7 MHz.
  3. Take the Real component of the result i.e. throw away the imaginary component.

For the upsampling process you have a couple of options. You can upsample by 12 and then decimate by 5. You would up sample by 12, apply the low-pass filter, and then down sample by 5. The filter would need a pass band from 0 - 4MHz (exact upper frequency depends on the nature of your IQ signal). The stop band would need to be at 5 MHz. You might be able to get away with a higher stopband frequency - e.g. if you have no frequency content between 4 MHz and 5 MHz, you could use a stopband starting at 6 MHz. The filter design is slightly dependent on the frequency analysis of your IQ signal and you would have to figure out your requirements for the filter specifications e.g. pass-band ripple and stopband attenuation.

For practical purposes you probably want to break the up sample by 12 into two separate steps e.g. Up sample by 4 and then up sample by 3 - each step would have its own filter applied.

An alternative method is to up sample by a factor of 3 or 4, apply the low-pass filter and then use a spline filter to interpolate the signal to required 24MHz sampling rate.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.