4 ms·

Each RF hop is going to have more latency than fiber for RF physical layer processing. Forward error correction intentionally uses large frames to spread out bu

by _flgt 7y ago

Each RF hop is going to have more latency than fiber for RF physical layer processing. Forward error correction intentionally uses large frames to spread out burst errors.

Diederich 7y ago
That's interesting, can you expand on that? Thanks.
paulsutter 7y ago
Sounds speculative. Given link bandwidths and likely error rates, it seems better to allow the higher layers to retransmit than to optimize for correcting burst errors. Especially given Starlink's goal to control latency.
- ianai 7y ago
  Have they given a range on latency goals? Ie seconds, tenths of seconds, milliseconds, etc? I’m seeing figures of 1 million$ per satellite. Over 10 years that’s roughly 100k/month in mortgage or around 1k customers at 100$/mo.
- _flgt 7y ago
  I haven't seen a link budget. Do you have a link for one about the "likely error rates"? My guess is Starlink is going to need FEC for Doppler effects and multipath in urban environments. Maybe they will be able to overcome multipath with modulation. But Doppler shift will be significant with LEO. The latest 802.11n standards have been increasing physical layer frame sizes, not decreasing them. I don't know of any modern wireless standard that doesn't use FEC. Coding gain is significant in most link budgets I have seen. The last satellite RF downlink I worked with was for JPSS-1 in LEO nearly 6 years ago. There was a (255, 223) Reed-Solomon code with an interleave depth of 4 frames. So you effectively had to receive 1024 bytes, decode, and you got 4x 223 bytes worth of information. While latency wasn't a major concern, this scheme seemed pretty standard to me. If you don't believe me, the spec is public information: https://www.star.nesdis.noaa.gov/jpss/documents/CDFCB/GSFC_474-00001-07-01_CDFCB_External_Vol.7-1_JPSS_Downlink_Data_Formats__D34862-07-01_.pdf https://www.star.nesdis.noaa.gov/jpss/documents/CDFCB/GSFC_4... I'm sure SpaceX is trying really hard to control latency - so it doesn't get out of hand! One set of RF processing here, another over there, each with their own filters and decoding algorithms and then you have an extra 100 ms of delay in the system.
  - paulsutter 7y ago
    Elon talks about 20ms or 10ms, not 100. He wouldn’t be wrong by a factor of 10. You’re really just speculating based on some non IP based overarchitected mil spec monstrosity you’ve seen before https://mobile.twitter.com/elonmusk/status/1132903914586529793?s=19 https://mobile.twitter.com/elonmusk/status/11329039145865297... Besides we’re talking about the latency of the satellite to satellite links, not the downlink (since that’s the difference between using a downlink close to the client vs close to the server).
    - _flgt 7y ago
      Yeah, I'm speculating and so are you. Your citation is an Elon tweet. I'm not talking satellite to satellite latency, I'm talking total latency end-to-end. I'm also not saying the system will be 100 ms, just that it takes significant engineering effort so latency doesn't get out of control. IMHO, it's a fundamentally much harder problem than existing transatlantic links, despite the greater signal speed in a vacuum. I'm skeptical of what the latency will be. You disagree and that's fine. Maybe we'll see an actual measurement soon enough. The competition can do 60 ms round-trip to NYC to London.
      - paulsutter 7y ago
        I think you’re suggesting an /extra/ 100ms in addition to speed of light time, which is to speculate that they’ll do it in a way that’s needlessly crappy. I’m speculating they’re smart guys working hard to keep latency low, and that routing will be a more challenging latency issue, while serialization should be minor. But we’ll see and I agree we’re both speculating
  - Tuna-Fish 7y ago
    Not commenting on error correction, but your speculation about multipath reveals that you fundamentally fail to understand how starlink is meant to work. Starlink allows high bandwidth density on the ground, and at the satellites because every antenna in the system is an extremely capable steerable phased array. Every transceiver can fully distinguish between signals of equal power and frequency that differ whose sources differ at least 10 degrees. Reflections from buildings are irrelevant when the antenna must already be capable of picking the correct transmission out of the 50 or so equally powerful beams on the same frequency that only differ in angle. Also, since every terminal already needs to be constantly aware of the position of every satellite, the correct answer to doppler is not fancy error correction, it's modulating the transmission to eliminate it, or just expecting it at the receiver.
wmf 7y ago
Tell that to the HFT people.
- _flgt 7y ago
  Is this some fixed point-to-point line of sight terrestrial microwave link? I'd believe that could do it. Do HFT people ever communicate with spacecraft though?
  - jandrese 7y ago
    No, the point of the HFT links was to make the most direct shot possible so they could beat the packets traveling via fiber. Going up to space would defeat the purpose.

by _flgt 7y ago

Each RF hop is going to have more latency than fiber for RF physical layer processing. Forward error correction intentionally uses large frames to spread out burst errors.

Diederich 7y ago

That's interesting, can you expand on that? Thanks.

paulsutter 7y ago

Sounds speculative. Given link bandwidths and likely error rates, it seems better to allow the higher layers to retransmit than to optimize for correcting burst errors. Especially given Starlink's goal to control latency.

ianai 7y ago
Have they given a range on latency goals? Ie seconds, tenths of seconds, milliseconds, etc? I’m seeing figures of 1 million$ per satellite. Over 10 years that’s roughly 100k/month in mortgage or around 1k customers at 100$/mo.
_flgt 7y ago
I haven't seen a link budget. Do you have a link for one about the "likely error rates"? My guess is Starlink is going to need FEC for Doppler effects and multipath in urban environments. Maybe they will be able to overcome multipath with modulation. But Doppler shift will be significant with LEO. The latest 802.11n standards have been increasing physical layer frame sizes, not decreasing them. I don't know of any modern wireless standard that doesn't use FEC. Coding gain is significant in most link budgets I have seen. The last satellite RF downlink I worked with was for JPSS-1 in LEO nearly 6 years ago. There was a (255, 223) Reed-Solomon code with an interleave depth of 4 frames. So you effectively had to receive 1024 bytes, decode, and you got 4x 223 bytes worth of information. While latency wasn't a major concern, this scheme seemed pretty standard to me. If you don't believe me, the spec is public information: https://www.star.nesdis.noaa.gov/jpss/documents/CDFCB/GSFC_474-00001-07-01_CDFCB_External_Vol.7-1_JPSS_Downlink_Data_Formats__D34862-07-01_.pdf https://www.star.nesdis.noaa.gov/jpss/documents/CDFCB/GSFC_4... I'm sure SpaceX is trying really hard to control latency - so it doesn't get out of hand! One set of RF processing here, another over there, each with their own filters and decoding algorithms and then you have an extra 100 ms of delay in the system.
- paulsutter 7y ago
  Elon talks about 20ms or 10ms, not 100. He wouldn’t be wrong by a factor of 10. You’re really just speculating based on some non IP based overarchitected mil spec monstrosity you’ve seen before https://mobile.twitter.com/elonmusk/status/1132903914586529793?s=19 https://mobile.twitter.com/elonmusk/status/11329039145865297... Besides we’re talking about the latency of the satellite to satellite links, not the downlink (since that’s the difference between using a downlink close to the client vs close to the server).
  - _flgt 7y ago
    Yeah, I'm speculating and so are you. Your citation is an Elon tweet. I'm not talking satellite to satellite latency, I'm talking total latency end-to-end. I'm also not saying the system will be 100 ms, just that it takes significant engineering effort so latency doesn't get out of control. IMHO, it's a fundamentally much harder problem than existing transatlantic links, despite the greater signal speed in a vacuum. I'm skeptical of what the latency will be. You disagree and that's fine. Maybe we'll see an actual measurement soon enough. The competition can do 60 ms round-trip to NYC to London.
    - paulsutter 7y ago
      I think you’re suggesting an /extra/ 100ms in addition to speed of light time, which is to speculate that they’ll do it in a way that’s needlessly crappy. I’m speculating they’re smart guys working hard to keep latency low, and that routing will be a more challenging latency issue, while serialization should be minor. But we’ll see and I agree we’re both speculating
- Tuna-Fish 7y ago
  Not commenting on error correction, but your speculation about multipath reveals that you fundamentally fail to understand how starlink is meant to work. Starlink allows high bandwidth density on the ground, and at the satellites because every antenna in the system is an extremely capable steerable phased array. Every transceiver can fully distinguish between signals of equal power and frequency that differ whose sources differ at least 10 degrees. Reflections from buildings are irrelevant when the antenna must already be capable of picking the correct transmission out of the 50 or so equally powerful beams on the same frequency that only differ in angle. Also, since every terminal already needs to be constantly aware of the position of every satellite, the correct answer to doppler is not fancy error correction, it's modulating the transmission to eliminate it, or just expecting it at the receiver.

wmf 7y ago

Tell that to the HFT people.

_flgt 7y ago
Is this some fixed point-to-point line of sight terrestrial microwave link? I'd believe that could do it. Do HFT people ever communicate with spacecraft though?
- jandrese 7y ago
  No, the point of the HFT links was to make the most direct shot possible so they could beat the packets traveling via fiber. Going up to space would defeat the purpose.