Service Data Unit Error Probability

Description

SDU error probability is a Quality of Service (QoS) metric defined in 3GPP TS 21.905 (Vocabulary for 3GPP Specifications) and referenced in TS 22.060 (General Packet Radio Service, GPRS). It quantifies the proportion of Service Data Units that are delivered to the destination with undetected errors, relative to the total number of SDUs that were either successfully transferred or found to be in error during a measurement sample. The definition is inherited from ITU-T Recommendation X.140, which provides a general framework for defining performance parameters for data communication networks.

A Service Data Unit is the unit of data that a given protocol layer receives from the layer above it and passes to the layer below after adding its own protocol control information. For example, an IP packet handed to a PDCP entity is an SDU from PDCP's perspective, while the PDCP PDU is what is actually transmitted over the radio interface. SDU error probability therefore measures errors at the boundary of the service offered by a layer to its upper layer, not at the physical transmission level. This distinction is crucial: it captures residual errors that survive all lower-layer protection mechanisms such as CRC checking, forward error correction, HARQ retransmissions, and ARQ/RLC retransmissions.

The metric is computed as the number of incorrect SDUs divided by the sum of successfully transferred SDUs and incorrect SDUs within a defined sample. An SDU is considered incorrect if it is delivered to the destination but contains bit errors that were not detected by the lower layers, or if it is delivered with an integrity failure. Note that lost SDUs are excluded from this metric — they are captured separately by SDU loss probability. This separation allows network operators and equipment vendors to distinguish between two fundamentally different failure modes: data that never arrives versus data that arrives corrupted.

In 3GPP systems, SDU error probability is typically specified as a target value in QoS profiles for conversational and streaming services. For instance, a conversational voice service might require an SDU error ratio of 10^-3, while a streaming video service might tolerate 10^-5 or better depending on the codec's error resilience. These targets drive the design of lower-layer error protection, including the choice of modulation and coding schemes, HARQ parameters, and RLC retransmission configurations. The metric is measured at the receiving end over a statistically significant sample of SDUs to provide a reliable estimate.

The term has remained stable across all 3GPP releases from Rel-4 through Rel-19, reflecting its foundational nature as a QoS parameter. While the underlying radio technologies have evolved dramatically — from GPRS through UMTS, LTE, and 5G NR — the conceptual definition of SDU error probability has not required revision because it operates at the service boundary and is technology-agnostic. This stability makes it a reliable cross-release reference point for QoS specification and comparison.

Purpose & Motivation

SDU error probability exists to provide a standardized, technology-independent way to specify and measure the residual error performance of a data delivery service. Before such metrics were formalized, QoS was often described in vague terms or in ways that were tightly coupled to specific transmission technologies, making it difficult to compare services, write interoperable specifications, or verify that a network met its performance commitments. By defining error performance at the SDU boundary — the interface between protocol layers — the metric abstracts away the details of how errors are detected and corrected underneath.

The metric solves a critical problem for network operators and service providers: it allows QoS guarantees to be expressed in terms that directly relate to application performance. An application does not care how many HARQ retransmissions occurred or what the block error rate was at the physical layer; it cares whether the data it receives is correct. SDU error probability bridges this gap by measuring exactly what the application experiences. This makes it possible to write meaningful service level agreements (SLAs) and to verify them through end-to-end measurements.

Historically, the concept was inherited from ITU-T X.140, which was developed for general data communication networks in an era when packet data services were emerging. 3GPP adopted and referenced this definition when specifying GPRS QoS in TS 22.060, ensuring alignment with international standards. The separation of error probability from loss probability was a deliberate design choice that recognizes that different applications have different sensitivities to corrupted versus missing data. For example, a streaming audio application may prefer to receive a corrupted packet and attempt concealment rather than lose the packet entirely and cause a gap, while a file transfer application would rather retransmit a lost packet than accept a corrupted one.