HSUPA vs HSDPA: Understanding 3.5G Architecture and Its Modern Relevance
A technical comparison of HSUPA (uplink) and HSDPA (downlink) within the HSPA framework, covering channel design, scheduling, HARQ, and how these mechanisms influenced LTE and 5G NR architecture.
HSDPA and HSUPA are frequently mentioned together under the HSPA umbrella, but they address different problems, use different channel structures, and were standardized in separate 3GPP releases. Understanding the distinction matters for anyone working with legacy UMTS networks, reading 3G performance data, or trying to understand why LTE and 5G NR made the architectural choices they did.
This article compares HSDPA and HSUPA at the channel, scheduling, and HARQ levels, then maps those design decisions to their equivalents in contemporary cellular systems.
Background: The WCDMA Baseline
Before either HSDPA or HSUPA, WCDMA (Wideband CDMA, Release 99) provided uplink and downlink data services through Dedicated Channels (DCH). Both directions used the same basic mechanism: RNC-controlled rate setting, soft handover on both uplink and downlink, and relatively slow adaptation to changing radio conditions.
The limitation was structural. The RNC sat above the Node B in the network hierarchy. Rate changes required signaling that traversed the Iub interface between Node B and RNC, introducing delays of 100 ms or more. In a radio environment where channel quality can change significantly within 10 ms, this reaction time severely limited throughput efficiency.
HSDPA (Release 5, 2002) addressed the downlink first. HSUPA (Release 6, 2004) followed for the uplink. Together they formed HSPA, High-Speed Packet Access, the system that would brand 3.5G networks commercially.
HSDPA: The Downlink Solution
HSDPA introduced the High-Speed Downlink Shared Channel (HS-DSCH). Key characteristics:
Shared channel, fast scheduling. HS-DSCH is time-shared among UEs in a cell on a 2 ms TTI basis. The Node B scheduler selects which UE to serve in each TTI based on channel quality reports (CQI) sent by UEs every 2 ms. This is proportional fair or maximum C/I scheduling, depending on implementation.
Adaptive modulation and coding (AMC). Rather than adjusting transmit power to compensate for channel variation (as DCH did), HSDPA adjusts modulation and coding rate. Good channel conditions: 16QAM with high-rate turbo code. Poor conditions: QPSK with low-rate code. The transmit power stays roughly constant; the information rate adapts.
HARQ at the Node B. Retransmissions are handled at the Node B rather than at the RNC. The UE sends ACK/NACK within one TTI, and the Node B retransmits immediately. Chase combining or incremental redundancy is used to combine retransmissions with previous attempts.
No soft handover. HS-DSCH does not support soft handover, which was a feature of DCH. Serving cell handover for HSDPA is handled by a hard handover mechanism (HS-DSCH serving cell change). This was initially considered a downgrade, but the gains from fast scheduling and AMC more than compensated.
HSDPA peak rates progressed through device categories:
| HSDPA Category | Modulation | Codes | Peak Rate |
|---|---|---|---|
| Cat 1-2 | QPSK | 5 | 1.2 Mbps |
| Cat 3-4 | QPSK | 5 | 1.8 Mbps |
| Cat 5-6 | QPSK / 16QAM | 5 | 3.6 Mbps |
| Cat 7-8 | 16QAM | 10 | 7.2 Mbps |
| Cat 10 | 16QAM | 15 | 14.4 Mbps |
HSUPA: The Uplink Solution
HSUPA, formally Enhanced Dedicated Channel (E-DCH), took a different approach than HSDPA because the uplink has fundamentally different constraints:
- Multiple UEs transmit simultaneously on shared spectrum, creating mutual interference.
- UE power is limited (typically 24 dBm maximum), unlike a base station.
- The UE, not the network, controls when and how much it transmits.
The E-DCH design addressed these constraints through Node B-controlled scheduling with UE feedback rather than network-unilateral scheduling as in HSDPA.
E-DCH Channel Structure
HSUPA introduced several new uplink channels:
E-DPDCH (Enhanced Dedicated Physical Data Channel): Carries user data. Up to four parallel E-DPDCHs can be configured for higher data rates (multi-code transmission).
E-DPCCH (Enhanced Dedicated Physical Control Channel): Carries control information from the UE: E-TFCI (transport format combination indicator, describing what the UE is transmitting), RSN (retransmission sequence number for HARQ), and Happy Bit.
E-AGCH (E-DCH Absolute Grant Channel): Downlink channel from Node B granting the UE an absolute maximum transmit power ratio. Used for initial grant and reconfiguration.
E-RGCH (E-DCH Relative Grant Channel): Downlink channel for incremental up/down commands. Allows the Node B to quickly adjust UE transmit power in steps without the overhead of a full absolute grant.
E-HICH (E-DCH HARQ Indicator Channel): Carries ACK/NACK for HSUPA retransmission control.
The Happy Bit
One of the more operationally interesting aspects of HSUPA is the Happy Bit, a single bit in the E-DPCCH that signals whether the UE is satisfied with its current grant. If the UE has more data buffered than its current grant allows it to transmit, it signals "unhappy" (0), prompting the Node B scheduler to consider increasing its grant. If the UE has no additional backlogged data, it signals "happy" (1).
This feedback mechanism allows the scheduler to prioritize UEs with pending data without requiring full buffer status reports in every TTI, an efficient encoding of scheduling-relevant information in a single bit. Later systems used more detailed Buffer Status Reports (BSR) in LTE and NR, trading simplicity for precision.
TTI Options and Scheduling
HSUPA supported two TTI lengths: 2 ms and 10 ms. The 2 ms TTI provided lower latency and faster HARQ feedback but required higher UE processing capability. The 10 ms TTI was specified first (Release 6) and was supported by a broader range of devices.
Most HSUPA deployments used 10 ms TTI initially. Carriers deploying HSUPA in conjunction with HSDPA in the same carriers found that 10 ms HSUPA was often sufficient to match realistic application uplink demands of the era.
Direct Comparison
| Dimension | HSDPA | HSUPA |
|---|---|---|
| 3GPP Release | Release 5 (2002) | Release 6 (2004) |
| Direction | Downlink | Uplink |
| Channel name | HS-DSCH | E-DCH |
| Scheduling entity | Node B | Node B (with UE feedback) |
| TTI | 2 ms | 2 ms or 10 ms |
| AMC | Yes (QPSK to 16QAM) | No (power variation + code rate) |
| HARQ | Chase combining / IR at Node B | Chase combining / IR at Node B |
| Soft handover | No (hard serving cell change) | Soft handover retained |
| Peak rate (Release 6) | 14.4 Mbps | 5.76 Mbps |
The asymmetry in peak rates reflects the fundamental difference between a base station transmitter (high power, multiple antennas, no size constraint) and a UE transmitter (power-limited, thermally constrained, form-factor limited). This asymmetry persists through LTE and into 5G, where downlink peak rates typically exceed uplink by a factor of 2-4x in standard configurations.
The retention of soft handover on HSUPA uplink was a notable difference from HSDPA. Because multiple cells receive the same uplink signal from a UE in soft handover, the signals can be combined at the RNC (macro diversity combining). This improves uplink coverage at cell edges, relevant for devices transmitting at power limits in poor coverage conditions.
Influence on LTE Design
LTE's architects borrowed from both HSDPA and HSUPA but simplified the structure considerably:
From HSDPA: The shared downlink channel model, Node B scheduling, AMC, and HARQ at the base station all carried forward to LTE. The PDSCH (Physical Downlink Shared Channel) is conceptually the HSDPA model matured.
From HSUPA: The Node B-controlled uplink scheduling model, HARQ on the uplink, and the principle of fast UE feedback carried forward to LTE's PUSCH (Physical Uplink Shared Channel). LTE eliminated the Happy Bit in favor of full Buffer Status Reports, and dropped soft handover entirely (LTE uses hard handover only), but the core scheduling loop originated with HSUPA.
What LTE discarded: The split RNC/Node B architecture was collapsed entirely. LTE's eNodeB handles all scheduling, HARQ, and handover decisions internally (with X2 interface coordination for handovers). This architectural change would have been much harder to make without the experience of deploying HSDPA and HSUPA, both of which demonstrated that moving intelligence to the base station worked.
Relevance for Modern Networks
HSPA networks remain operational in many markets as 4G and 5G coverage continues to expand. Some specific contexts where HSUPA/HSDPA behavior remains practically relevant:
IoT fallback scenarios. NB-IoT and LTE-M devices are designed to fall back to GPRS in areas without LTE coverage. In regions where HSPA is the highest available technology, IoT devices may operate on HSPA rather than LTE. Understanding HSUPA E-DCH behavior is relevant to troubleshooting uplink performance in these scenarios.
Legacy MVNO infrastructure. Several MVNOs continue to offer HSPA-only plans on roaming agreements. Devices operating on these plans see HSUPA uplink performance, and diagnostic tools that report E-DCH category information can help distinguish between HSPA device category limitations and coverage limitations.
3G sunset planning. Network operators decommissioning 3G infrastructure need to understand which traffic types depended on HSUPA characteristics, specifically, applications that relied on sustained uplink performance at sub-5 Mbps rates. Migrating these to LTE requires ensuring adequate uplink coverage in the areas previously served by HSPA.
Conclusion
HSDPA and HSUPA solved complementary problems using compatible mechanisms, both moved scheduling to the Node B, both implemented HARQ at the air interface level, and both used short TTIs to enable fast adaptation. Their differences, HSDPA's downlink shared channel versus HSUPA's uplink scheduled grant model, HSDPA's AMC versus HSUPA's power and code adaptation, reflect the distinct physical constraints of the two link directions.
Those architectural decisions were not discarded when LTE arrived; they were simplified, unified, and scaled. Understanding the original HSPA design makes the LTE and 5G NR air interface considerably more legible.
Related reading on hsupa.com: