NTN  

 

HARQ in NTN

A satellite link can have good signal quality and still fail to keep the data rate high. One reason is HARQ process availability. The transmitter may have used every available process while it is still waiting for the result of the earliest transmission.

NR NTN keeps the HARQ framework but adds ways to handle the long round trip. More processes provide more parallel work. Downlink feedback disabling and uplink HARQ mode B allow different scheduling behavior when waiting for a full RTT becomes expensive.

This page focuses on NR NTN unicast HARQ. NB-IoT and LTE NTN have their own procedures and parameters. For the slot relationships behind the transmissions, see Scheduling & Timing.

Why Long RTT Occupies HARQ Processes

Think of a HARQ process as the state for a transport block and its possible retransmissions. With feedback-driven operation, the transmitter needs time to learn the decoding result and decide what to schedule next. Other processes allow transmission to continue while that result is pending.

The relevant interval includes more than propagation alone. Feedback timing, receiver processing and scheduler behavior also contribute. A satellite orbit name by itself does not give the complete process reuse time.

If all available processes are occupied, this UE can become process-limited even when more radio resources are available. The gNB may still schedule other UEs. A pause for one UE is not necessarily an idle cell.

Four processes fill four slots. In this simplified one-UE example, the next four slots cannot carry a new block for that UE. Process zero is reusable at eight milliseconds.Four processes fill four slots. In this simplified one-UE example, the next four slots cannot carry a new block for that UE. Process zero is reusable at eight milliseconds. Illustrative model: 4 processes, 1 ms slots, process reuse after 8 ms 0 ms P0 1 ms P1 2 ms P2 3 ms P3 4 ms Waiting 5 ms Waiting 6 ms Waiting 7 ms Waiting 8 ms P0 0-4 ms: four different processes carry four new transport blocks. 4-8 ms: all four are occupied; this UE has no free process in the model. 8 ms: process 0 becomes available again. Other UEs may use the earlier gap. Model utilization = 4 processes x 1 ms / 8 ms = 50%

Figure 1. More propagation and feedback delay can occupy every process before the first one becomes reusable.

A Simple Process-Availability Calculation

Assume one new transport block is scheduled per slot, with one process occupied for each block. Let T_reuse be the time until that process is available again. This simplified model excludes retransmissions, repetitions and multi-slot transmissions.

Required processes ≥ ceil(T_reuse / T_slot)

Scheduling utilization ≤ min(1, N_processes x T_slot / T_reuse)

The second expression is a process-availability bound for this model. It is not an end-to-end throughput formula. Transport block size, coding rate, PRBs, layers and protocol overhead still determine the delivered bit rate.

These examples use invented reuse intervals to isolate the arithmetic. At 15 kHz SCS a slot is 1 ms; at 30 kHz it is 0.5 ms.

T_reuse

SCS / slot duration

Processes

Processes needed for one TB per slot

Model utilization bound

24 ms

15 kHz / 1 ms

16

24

66.7%

24 ms

15 kHz / 1 ms

32

24

100%

24 ms

30 kHz / 0.5 ms

32

48

66.7%

500 ms

15 kHz / 1 ms

32

500

6.4%

The 30 kHz example is worth noticing. The propagation time did not increase, but the scheduler has twice as many slot opportunities per millisecond. The same number of processes therefore covers fewer milliseconds at one new block per slot.

Adding processes helps within a finite range. It does not remove propagation delay, and it also requires more parallel HARQ state and buffering.

Configuring the Process Count

NR NTN supports configuration of 32 downlink HARQ processes through nrofHARQ-ProcessesForPDSCH-v1700. Uplink has nrofHARQ-ProcessesForPUSCH-r17 for the 32-process configuration. These are separate settings for separate directions.

Do not assume that every NTN UE always operates with 32 processes. Check capability support, the received configuration and the DCI format. Process-number interpretation must agree between UE and gNB.

For PDSCH, TS 38.331 specifies eight processes when both nrofHARQ-ProcessesForPDSCH and nrofHARQ-ProcessesForPDSCH-v1700 are absent. The default is therefore not automatically 16 or 32. See PDSCH-ServingCellConfig.

Downlink HARQ Feedback Enabled or Disabled

For downlink, UE normally returns HARQ-ACK information for received PDSCH transport blocks. NTN allows the network to disable this feedback per HARQ process. This is controlled by downlinkHARQ-FeedbackDisabled-r17 in PDSCH-ServingCellConfig.

Disabling feedback allows a process to be scheduled again before one HARQ RTT has elapsed since its previous scheduling. The gNB does not have to wait for the usual feedback round trip for that process. Normal scheduling and receiver-processing requirements still apply.

The setting removes the corresponding over-the-air HARQ feedback obligation. It does not remove channel coding, CRC checking, HARQ process identities or all possible retransmission handling. It also does not convert every failed transport block into successfully delivered data.

The scheduler can use suitably configured redundancy or repetitions, while higher layers can recover losses where their protocols support recovery. RLC AM can retransmit missing data; RLC UM does not provide ARQ recovery. Higher-layer recovery also pays the satellite delay.

A missing ACK in a trace therefore needs configuration context. It could be expected for a feedback-disabled process, or it could indicate an actual failure on a feedback-enabled process.

Uplink HARQ Mode A and Mode B

Uplink uses a different control model. UE sends PUSCH, and the network controls subsequent transmissions through uplink grants and HARQ state. Do not look for an LTE PHICH-style ACK channel in NR.

The network configures uplinkHARQ-mode-r17 per process in PUSCH-ServingCellConfig. Mode B allows scheduling a process before one HARQ RTT has elapsed since it was last scheduled. Mode A retains the corresponding RTT-based behavior.

Mode B is not the same field as downlink feedback disabling. It does not give UE permission to transmit without resources or to ignore the HARQ process ID, NDI or RV. The applicable MAC procedure still determines whether a transmission contains new data or retransmitted data.

When configured grants are used, the network must ensure a suitable HARQ-mode configuration for the processes used by that grant. The equivalent consistency issue exists for downlink SPS and its feedback configuration. See TS 38.300, clause 16.14.2.1.

Reading the Two HARQ Bitmaps

Both configurations use a 32-bit string, but the meaning of a one bit is different. In each string, the first/leftmost bit represents process ID 0. Bits for process IDs that are not configured are ignored.

The examples below describe the first four bits of an otherwise explicitly configured 32-bit string. They are explanatory prefixes, not complete ASN.1 values.

Configuration

Bit 1

Bit 0

Prefix 1100 means

downlinkHARQ-FeedbackDisabled-r17

Feedback disabled

Feedback enabled

DL process 0 and 1: disabled; process 2 and 3: enabled.

uplinkHARQ-mode-r17

Mode A

Mode B

UL process 0 and 1: mode A; process 2 and 3: mode B.

For example, all-zero downlink configuration enables feedback for the configured processes. An all-zero uplink mode bitmap selects mode B. Copying a bitmap between the two settings would therefore change its meaning.

These bit assignments are specified in TS 38.331. The NTN RRC reference provides the surrounding parameter structures.

HARQ Timing and DRX

A long propagation delay also affects when UE should expect a retransmission assignment or grant. A short terrestrial waiting period can make UE monitor too early or interpret the expected response interval incorrectly.

TS 38.321 defines HARQ-RTT-TimerDL-NTN for applicable feedback-enabled downlink processes and HARQ-RTT-TimerUL-NTN for applicable mode-A uplink processes. Under the specified DRX conditions, they add the latest UE-gNB RTT to the corresponding configured DRX HARQ RTT timer.

The MAC definition of UE-gNB RTT combines UE Timing Advance and kmac, expressed in compatible time units. This is another reason not to replace all timing terms with one generic satellite RTT.

Check the exact timer start event, including the relevant feedback transmission or PUSCH transmission. A DRX HARQ RTT timer is not the same thing as the overall T_reuse used in the simplified calculation above. See TS 38.321, clause 5.7.

What Changes When Feedback is Disabled ?

Feedback-driven retransmission gives the network a decoding result, but the result takes time to return. Removing that dependency improves scheduling freedom while reducing the immediate information available about successful delivery.

The choices below describe engineering trade-offs. The best configuration depends on radio errors, traffic requirements, delay and implementation support; there is no universal rule that every GEO or LEO deployment uses one mode.

Choice

Helps with

Still needs attention

More processes

Keeping transmissions in flight while results are pending.

Buffering and finite process count.

DL feedback disabled

Reusing processes without waiting for the feedback RTT.

Loss recovery, redundancy and actual delivered throughput.

UL mode B

Scheduling before a full HARQ RTT has elapsed.

Grant handling, HARQ state and configured-grant consistency.

More robust transmission

Reducing the probability of a failed decode.

Resource cost and achievable data rate.

Measure useful delivered bytes rather than scheduled transport blocks alone. A busy transmitter can still deliver poor application throughput if losses repeatedly reach a slow recovery layer.

What to Check in a HARQ Trace

First separate downlink and uplink. Then follow each process ID over time, together with the configured mode and the applicable scheduling offsets. A repeated process ID is not enough to identify a retransmission.

Use these checkpoints to distinguish a process bottleneck from a radio decoding or timing problem.

Checkpoint

Check

Configuration

Configured process counts, downlink bitmap, uplink mode bitmap and UE capabilities.

New data / retransmission

HARQ process ID, NDI transitions, RV and the applicable MAC rules.

Timing

Actual scheduled slots, K_offset, TA and expected feedback or grant timing.

Process availability

Whether all processes are occupied when data remains queued.

DRX

NTN RTT timers, active time and retransmission monitoring.

Reliability

CRC failures, RLC recovery where enabled, packet gaps and delivered throughput.

For the grant and ACK slot calculations, continue with Scheduling & Timing. For signal arrival alignment, see Timing Advance.

References

The explanations use these Release 18 versions. Numerical examples and diagrams are illustrative calculations, not measured results or universal satellite delay values.