To investigate whether the proposed Starlink Gen2 constellation will be competitive with "shortwave" communications systems, we've analyzed the proposed constellation plans, filed https://fcc.report/IBFS/SAT-LOA-20200526-00055/2378669.pdf.
To get a rough estimate, following pessimistic assumptions were made:
Satellites within each orbital plane are presumed to be able to have inter-satellite links to any other satellite within same plane, up to 6000km distance (data from ISL manufacturer). This is a reasonable assumption, as satellites in same orbital plane have minimal relative motion.
As actual distribution of satellites within the plane is unknown, random distribution is presumed (both RAAN and mean anomaly). This is a very pessimistic assumption, as any other distribution would have satellites more evenly spaced.
Orbits at 360km, 604km and 614km are ignored (greyed out above), as ability to cross-link between multiple orbital planes at same altitude is complicated.
No "ground stations" (other than endpoints) can be used for data transmission.
No connectivity between satellites at different orbital planes.
Only "distance-based" latency is considered - not uplink/downlink modulation latency. ISL serialization latency is presumably negligible.
An important parameter is the minimal elevation angle between ground and satellite - the lower the angle, the lower is possible latency. Starlink filings list minimum elevation angle of 40° for its user terminals, and 25° for earth gateways. Optimistic assumption is made that there is a gateway directly at CME and LD4 datacenter (i.e. communication is not done with an ad-hoc user terminal).
For Chicago-London connectivity, the lowest orbits (328km and 334km) are not suitable, as they have insufficient inclination to be useful for London. Three orbital planes remain: 345km, 373km and 499km.
For ease of comparison, numbers are shown as "latency in excess of great circle" (speed of light on a 6382km path over the surface of Earth).
To have an understanding of progressive improvement of latency as more satellites are launched, calculations are done for constellations in various stages of completion.
Because of the motion of satellites, latency is a probabilistic number, and can be described by a "cumulative density" graph: Y axis of represents probability that "excess latency" is below the corresponding point on the X axis.
For reference, for a "shortwave" (HF radio) transmissions, the "excess latency" varies between 0.2 to 1.2ms, caused by:
50-250km "effective height" of ionosphere (not identical to the hoF2)
curved path through ionosphere
ionization delay (about 0.1ms). Ionization delay is inversely proportional to square of frequency, and is negligible for Starlink.
Of course, "shortwave" transmissions also incur high "demodulation/decoding" delay, not included above - and are limited to small number of "bits" (trading signals).
Some conclusions can be drawn:
At constellation completion of only 5% (1500 satellites out of 30000 planned), Starlink excess latency is <4ms, 60% of time. This is likely to happen before the end of 2021, and is likely to beat total latency of most shortwave systems
At constellation completion of 50%, excess latency is below 2ms, 70% of time. This is likely to beat almost every shortwave system - except, of course, for ours.
For additional information about shortwave use for trading, please see our website, and posts by Sniper in Mahwah and Bob Van Valzah:
Previous research on this subject was based on the 2018 "Gen1" constellation plans, by Mark Handley at UCL: https://discovery.ucl.ac.uk/id/eprint/10062262/7/Handley_hotnets.pdf and https://www.youtube.com/watch?v=QEIUdMiColU - Gen2 is much more dense and at much lower orbits.