The day when starting a receiver fixed the transmitter | Hacker News Reader

The day when starting a receiver fixed the transmitter

rachelbythebay.com

89 pointsingve13 comments

> I recorded about 10 seconds of the ENTIRE FM band to a ramdisk, then turned around and pushed it back out in the unlicensed band on a continuous loop.

> Now I had a second problem on my hands: was it the recording itself? To answer that question, I took the file and washed it through a bunch more SDR gunk to isolate a single station, demodulate it to audio, and pushed it to my machine's speakers. The song sounded fine. Other stations were also okay.

Rachel, I would pay double-digit dollars per month for a screencast per week of you doing some serious hacker shit like this. (I've used Gnuradio et al and it's always a screaming pain, at least for a novice like me.) Watching someone hack away at it in a while(!functioning) would be pure pleasure, and I'm sure quite educational too.

I'm not Rachel and this is not a podcast, but I have wanted to write a similar tutorial for a while. So I finally did it: https://www.abclinuxu.cz/blog/jenda/2019/11/gnu-radio-first-...

I really enjoy reading these posts from Rachel. I think I have read like 10+ of them by now.

Can I be so bold as to suggest something for improved readability? [Edit: turns out I cannot. Immediate downvote!]

Breaking that wall-of-text into 2-3 sections with headlines would probably improve readability greatly. Honestly, it doesn't matter that much what the headlines say - the important thing is breaking up that wall of text into separate pieces.

> I recorded about 10 seconds of the ENTIRE FM band to a ramdisk

We live in wonderful age of SDR where this is not only possible but apparently also the way how even entry-level car infotainment's FM receiver works. And it still amazes me.

Was there really a time where HDDs were to slow for 40MB/s but we still had 400MB of RAM? Did I miss a factor of 2 because of the I/Q?

Depends on the box, but some years ago, I picked up an old slotted-Xeon server at a swap for a few bucks, and it had 32 DIMM slots on an enormous memory sub-board. Took registered or buffered ECC, as I recall. I scored a few modules on eBay and put 256MB into it, I think it would've gone to 4GB if I filled all the slots.

But the hard drives were SCSI SCA-2, which I believe maxed out at 40MB/s but in practice the drives were much slower. So yes, there was probably a point in time where a reasonably-beefy box would have had the specs you describe.

20 MHz --> 40M samples per second (Nyquist). 16 bits per channel, and two channels (I, Q).

40000000 x 2 x 2, so... about 160 MB/sec?

It's a lot of data. It wasn't happening with the disk I had on hand, so I went for the ramdisk thing. 1.6 GB of memory? That I did have.

I think you have an extra factor of two there. If you need 20 MHz without aliasing, you either need real sampling at 40 MS/s or IQ sampling at 20 MS/s, but IQ sampling at 40 MS/s is overkill.

Nyquist works differently if your samples are complex.

20 MHz only requires 20 Msps of complex samples. It would be 40 Msps if those were real samples.

Would be nice if you could use flowgraphs to program any fpgas on the radio.

You can with RFNoc blocks in GNU Radio, but that’s specific to Ettus hardware.

I had a P3 desktop with a UDMA hard drive (33MB/s theoretical peak) and 128MB of RAM. A workstation of the time would have over 512MB of RAM and a SCSI drive that might not sustain 40MB/s.

Going back further, in 1995, a SPARCstation 20 had a SCSI-2 controller (10MB/s maximum transfer rate) and could be fitted with 512MB of RAM.

I mean SATA 3 has a bit more than 500MB/s right, and PCIe SSDs run at between 1-4 GB/s? so a machine today (at least 8GB of RAM maybe in the hundreds) would have an even bigger ratio between memory size and disk transfer speed.

Reply on news.ycombinator.com