Service Data Unit Loss Probability

Description

SDU loss probability is a Quality of Service (QoS) metric defined in 3GPP TS 21.905 and referenced in TS 22.060 for GPRS. It measures the proportion of Service Data Units that are transmitted but never successfully delivered to the intended receiver, relative to the total number of SDUs transmitted during a measurement period. The definition originates from ITU-T Recommendation X.140, which provides a standardized framework for data communication performance parameters.

A Service Data Unit represents the data payload that a protocol layer receives from the layer above and encapsulates into a Protocol Data Unit (PDU) for transmission. When an SDU is lost, it means the receiver never obtains the data, even after all lower-layer recovery mechanisms — such as HARQ retransmissions at the physical layer and ARQ retransmissions at the RLC layer — have been exhausted or have timed out. The loss can occur due to unrecoverable transmission errors, buffer overflow and subsequent discard, excessive delay causing the SDU to be dropped, or network congestion leading to queue management discards.

The metric is calculated as the number of lost SDUs divided by the total number of transmitted SDUs in the sample. Unlike SDU error probability, which measures corrupted-but-delivered data, SDU loss probability captures data that is entirely absent at the receiver. This distinction is important because the two failure modes have different impacts on applications: lost data typically requires retransmission at higher layers (e.g., TCP) or results in permanent gaps in the received stream, while corrupted data may be usable if the application can tolerate or conceal errors.

In 3GPP QoS frameworks, SDU loss probability is specified as a target for various service classes. Conversational services like voice over IP typically require a loss probability of 10^-2 or better, while streaming services may target 10^-3 to 10^-5 depending on the codec's ability to handle gaps. Background and interactive services may tolerate higher loss rates, especially if TCP is used at the transport layer to recover lost segments. These targets influence radio resource management decisions, including scheduling priorities, retransmission limits, and admission control thresholds.

SDU loss probability is measured at the receiving endpoint over a statistically significant number of SDUs to provide a reliable estimate. In practice, measurement may be performed at various protocol layers depending on the scope of the QoS guarantee. For example, loss measured at the PDCP layer captures losses over the radio interface, while loss measured at the application layer captures end-to-end losses including transport network effects. The 3GPP definition is deliberately layer-agnostic, allowing it to be applied at whichever service boundary is relevant to the QoS specification. The term has remained unchanged across all releases from Rel-4 through Rel-19.

Purpose & Motivation

SDU loss probability exists to provide a precise, standardized metric for quantifying packet loss in 3GPP networks. Packet loss is one of the most fundamental impairments affecting data services, directly impacting application quality for voice, video, and data applications. Without a standardized definition, different vendors and operators might measure loss differently — some counting packets that arrive too late as lost, others counting only packets that never arrive — leading to inconsistent QoS reporting and inability to compare performance across networks or verify SLA compliance.

The metric addresses the need to separate loss from error in QoS specifications. In early packet data networks, performance was often described simply in terms of bit error rate or frame error rate, which conflated corrupted data with missing data. ITU-T X.140 recognized that these are distinct phenomena requiring separate metrics, and 3GPP adopted this separation when defining GPRS QoS in TS 22.060. This allows network designers to independently optimize error detection/correction mechanisms (which affect error probability) and resource allocation/retransmission strategies (which affect loss probability).

Historically, the need for this metric grew with the proliferation of packet-switched services that could not tolerate the loss characteristics of early radio systems. Circuit-switched voice had well-understood quality metrics, but packet data introduced new challenges: variable delay, bursty loss patterns, and the need to support diverse applications with different loss tolerances. By defining SDU loss probability in a technology-independent way, 3GPP enabled consistent QoS specification across GPRS, UMTS, LTE, and 5G NR, even as the underlying radio technologies changed dramatically. The metric remains relevant in 5G because applications still experience packet loss, and operators still need to specify and measure it consistently.