One thing that seems missing to me is that while the probe might send 160 bits/sec of useful data, those bits are not sent directly as such[1]:

The TMU encodes the high rate data stream with a convolutional code having constraint length of 7 and a symbol rate equal to twice the bit rate (k=7, r=1/2).

So the effective symbol rate is 320 baud[2], and thus a factor of two should be included in the calculations from what I can gather.

Note that the error correction was changed after Jupiter to use Reed-Solomon[3] (255,223) to lower the effective bit error rate, so effectively I guess the data rate is more like 140 bps.

[1]: https://web.archive.org/web/20130215195832/http://descanso.j...

[2]: https://destevez.net/2021/09/decoding-voyager-1/

[3]: https://destevez.net/2021/12/voyager-1-and-reed-solomon/

It looks like a pretty reasonable order of magnitude estimate to me. Energy arguments tend to be quite neat for that because if the efficiency is at all reasonable they constrain things well with fairly simple calculations. The antenna directionality is also reasonably well understood and characterised. The exact noise level discussed later on is probably where things get a bit more uncertain (but aren't directly needed to answer the question).

I was surprised, too. Knowing little about physics, this was a pleasant surprise, however.