What does good D2C performance mean? A short message, a map download and a voice call need different things. A peak download rate from one demonstration cannot describe all three services.

This page defines useful measurements for conventional D2C and works through simple calculations. The numbers are explanatory assumptions, not commercial service promises. For radio mechanisms, see Radio.

• Separate Coverage, Availability and Performance
• Worked Example: Sharing a Beam
• Worked Example: Propagation and Application Delay
• Worked Example: Intermittent Service
• Measure Failed Attempts Too
  • Why Voice Is a Different Test
• A Repeatable Performance Test
• References

Separate Coverage, Availability and Performance

Coverage describes where a usable link can exist under stated conditions. Availability adds the time dimension. Performance describes what a service delivers when it is attempted, including failures and periods without connectivity.

A coverage map may describe land area rather than population, and may assume an unobstructed sky. An availability claim for a cell is not automatically availability for a user whose antenna is blocked. State both the network conditions and the device conditions.

Worked Example: Sharing a Beam

Assume one beam has 5 MHz of usable downlink bandwidth. For this simplified calculation, average spectral efficiency before overhead is 1 bit/s/Hz and 30% of the resulting rate is lost to overhead and other non-payload use.

The modeled aggregate useful rate is 5,000,000 x 1 x 0.70 = 3.5 Mbit/s. Equal sharing gives 350 kbit/s for 10 simultaneous users, 70 kbit/s for 50, and 35 kbit/s for 100.

Figure 1 : More simultaneous users reduce the arithmetic per-user share

Real scheduling is not necessarily equal, and weak users may consume more resources per delivered bit. Retransmissions, traffic bursts and priority policy change the result. Also, a satellite can have multiple beams, but their resources may share power, processing or feeder capacity.

Worked Example: Propagation and Application Delay

Assume a 1,200 km service link and a 1,800 km feeder link with the radio path relayed to the ground. Ignoring all processing, the one-way propagation time is (1200 + 1800)/300000 = 10 ms, or 20 ms round trip.

Now assume the full request-response transaction adds 15 ms of scheduling and processing, 25 ms of transport/core delay, 30 ms of queuing and 10 ms of server work. If these non-overlapping components sum as stated, application latency is 20 + 15 + 25 + 30 + 10 = 100 ms.

This is an accounting example. Some operations overlap, some requests require several round trips, and a different route changes the propagation component. Specify whether a quoted latency is one-way, RTT, time to first byte or complete transaction time.

Worked Example: Intermittent Service

Suppose usable service lasts 6 minutes in each repeating 10-minute interval at a fixed location. The simplified availability is 6/10 = 60%, with a 4-minute gap between service intervals.

For uniformly distributed arrival times and a service that can begin immediately during coverage, 40% of requests arrive in a gap. Their mean wait is 2 minutes. The unconditional mean coverage wait is therefore 0.4 x 2 = 0.8 minutes, or 48 seconds.

This model excludes acquisition time and assumes each transaction is short enough to finish within the remaining service window. A large download can span multiple windows, so its completion time needs a different model. Average availability alone does not describe the longest outage.

Measure Failed Attempts Too

Suppose 200 messages are attempted during a test and 190 arrive within 60 seconds. Delivery within that deadline is 95%. If another five arrive after two minutes, they improve eventual delivery but do not change the 60-second result.

Report latency percentiles with their population. A median computed from successful attempts alone excludes the timeouts. Pair it with the attempt count, delivery ratio, timeout duration and longest observed delay.

Why Voice Is a Different Test

Voice needs repeated timely packet delivery rather than one eventual successful payload. Delay variation, loss bursts and coverage interruptions can damage a call even when the average data rate appears sufficient.

Measure call setup success, interruption duration and sustained quality over the intended service interval. Distinguish native cellular voice from a supported application call. A service that offers messaging does not automatically offer either form of voice.

A Repeatable Performance Test

Use a test matrix that changes one major condition at a time. Include several locations and repeat across time, because satellite geometry and network load can change between two otherwise identical attempts.

A useful report includes raw timestamps and failure causes alongside summary figures. Clearly label a controlled demonstration, a commercial field test and a model. The published September 2023 demonstration is an example of a dated test result; it is not a universal rate guarantee. See Deployments for evidence categories.

References

• AST SpaceMobile - September 2023 connectivity demonstration : Dated unmodified-handset test; separate voice and data results.
• Virgin Media O2 - O2 Satellite launch, 26 February 2026 : Operator announcement of messaging and supported-app data service.