Overview — Completing the Uplink

By the time Rel-6 was frozen in 2004, HSDPA was being rolled out by early adopter operators, delivering dramatically better downlink performance. But the uplink remained at the basic UMTS level: up to 384 kbps in good conditions, managed slowly by the RNC. For data applications that need to push content upward — uploading photos, video calls, email with attachments — the asymmetry was becoming a bottleneck.

Rel-6's answer was HSUPA (High Speed Uplink Packet Access), formally specified as the Enhanced Uplink (EUL) feature in 3GPP documents. HSUPA applies to the uplink the same two innovations that made HSDPA work on the downlink: moving scheduling into the Node B, and introducing fast HARQ with soft combining at the physical layer.

Alongside HSUPA, Rel-6 introduced MBMS — a way to broadcast the same data stream to many users simultaneously over the radio — and PoC (Push-to-Talk over Cellular), a walkie-talkie service built on IMS that targeted enterprise and public safety users.

HSUPA — Fast Uplink with Node B Scheduling

HSUPA introduces the E-DCH (Enhanced Dedicated Channel)as its uplink transport channel, replacing or augmenting the old DPDCH/DPCCH channel pair for packet data. Like HSDPA's HS-DSCH on the downlink, the E-DCH is designed to carry short bursts of data with very fast scheduling and retransmission.

Uplink Scheduling at the Node B

In basic UMTS, the RNC allocated uplink resources (spreading factor, maximum power) to each UE for the duration of a connection, adjusted slowly based on congestion measurements. With HSUPA, the Node B runs the uplink scheduler: the UE sends a scheduling request indicating how much data it has to send and what channel conditions look like; the Node B responds with a grant — an absolute grant or a relative adjustment — within a few milliseconds. The UE then transmits on the E-DCH within the granted resource.

Two channels support the E-DCH on the uplink:

Fast HARQ on the Uplink

HSUPA uses the same HARQ principle as HSDPA, but operating in the reverse direction:

• Chase combining — the Node B stores soft bits from failed uplink transmissions; retransmissions are identical copies that get combined with the stored version, adding signal energy each time
• Incremental redundancy (IR) — retransmissions send different redundancy bits rather than identical copies, so each attempt adds new coding information that collectively reduces the effective code rate
• The NACK-to-retransmit round trip is handled entirely within the Node B and UE, with no RNC involvement, completing in approximately 10 ms for the 2 ms TTI variant

Peak uplink: 5.76 Mbps (Category 6), up from 384 kbps in basic UMTS. A 15× improvement on the uplink, matching the scale of HSDPA's gains on the downlink.

HSPA = HSDPA + HSUPA — Mobile Broadband Complete

With HSDPA (Rel-5 downlink) and HSUPA (Rel-6 uplink) both standardised, the industry coined a marketing term that captured the combination: HSPA, also marketed as 3.5G. The full numbers:

• Downlink: up to 14.4 Mbps (HSDPA Cat 12, with 16-QAM)
• Uplink: up to 5.76 Mbps (HSUPA Cat 6, with HARQ)
• Round-trip latency: typically 50–70 ms, versus 150–300 ms for basic UMTS — a decisive improvement for interactive applications

This is the capability set that the first-generation iPhone did not have at launch — Apple shipped the original 2007 model with EDGE (2.5G) and added HSPA in the iPhone 3G (2008). The smartphone data explosion that followed, from 2008 to 2012, ran almost entirely on HSPA networks. HSPA handled the majority of mobile data traffic worldwide until LTE deployments became widespread from 2013 onward.

MBMS — One Radio Bearer for Many Receivers

MBMS (Multimedia Broadcast Multicast Service) addresses a fundamental inefficiency in cellular unicast delivery: when many users in the same cell want the same content simultaneously, the network carries a separate copy for each of them. In a stadium or a busy city street during a major event, this can collapse a cell.

MBMS defines a broadcast mode where a single radio bearer is transmitted by the base station and received simultaneously by all devices tuned to that bearer — regardless of how many receivers there are. The radio resource consumed is constant: one bearer, one chunk of bandwidth, whether 10 or 10,000 users are watching.

How MBMS Works

• The network establishes an MBMS bearer service — a one-to-many radio channel — for a specific content stream (e.g. a live football broadcast)
• The base station broadcasts the MBMS data on a dedicated shared channel; UEs interested in the service tune their receivers to that channel
• Multiple cells can broadcast the same MBMS content simultaneously, allowing UEs to receive from the strongest cell without a handover interruption
• A Broadcast Multicast Service Centre (BM-SC) manages content sourcing, authorisation, and the MBMS session announcement

MBMS Applications

MBMS was designed for content that many users want at the same time and place:

• Live sports broadcasts and breaking news streams during peak demand
• Emergency alerts and civil warning messages sent simultaneously to all devices in an area
• Large software or firmware updates pushed to fleets of devices simultaneously
• Digital terrestrial television delivered over the cellular network

MBMS in Rel-6 was designed for UMTS networks; 3GPP later refined it as eMBMS (evolved MBMS) for LTE and further as MBS (Multicast Broadcast Services) for 5G NR — the same fundamental architecture, progressively enhanced.

Push-to-Talk over Cellular (PoC)

PoC (Push-to-Talk over Cellular) brings the walkie-talkie model to a mobile network. Unlike a phone call — where both parties have dedicated two-way audio simultaneously — PoC is half-duplex: only one group member transmits at a time, and all other members receive.

The PoC session model works as follows:

• A user presses and holds a button on their device, sending a floor request to the PoC server
• The PoC server grants the floor — giving that user the right to transmit — within a few hundred milliseconds (the "talk burst" setup)
• The user speaks; the audio is encoded and sent as RTP packets over the PS domain to the PoC server
• The PoC server distributes the audio simultaneously to all other group members, who receive it in near-real-time
• When the user releases the button, the floor is released and any other group member can then request it

PoC uses IMS (standardised in Rel-5) as its session layer — SIP for group session management and floor control. The media itself runs as RTP over the PS domain, reusing the existing packet core infrastructure. No dedicated circuit resources are consumed.

Target markets included enterprise field workers (utilities, construction, logistics), emergency services, and public safety teams who needed group communications without the setup time of a phone call. Modern PTT (Push-to-Talk) services — including FirstNet in the United States, built on LTE — are direct descendants of the PoC architecture defined in Rel-6.

WLAN Interworking

By 2004, dual-mode handsets capable of both 3G cellular and 802.11 Wi-Fi were emerging. Rel-6 began the work of standardising how a UE discovers, selects, and attaches to a Wi-Fi access point while maintaining its 3G identity and services — the first step toward the seamless cellular/Wi-Fi switching that all modern smartphones perform automatically.

Rel-6 defined several interworking scenarios, ranging from loose coupling (the UE uses Wi-Fi for internet access independently, with no interaction with the 3GPP core) to tighter coupling (the Wi-Fi access point authenticates using the SIM card via EAP-SIM, giving the network visibility of the subscriber on Wi-Fi). The concept of the ANDSF (Access Network Discovery and Selection Function) — a network entity that provides the UE with policies for when to prefer Wi-Fi over cellular — was studied during Rel-6 and formally standardised in later releases.

This work established the principle that a subscriber's SIM identity and associated policies should follow them seamlessly across access technologies — a principle that became central to 4G and 5G architecture.