Random Access (Contention & Non-Contention) in LTE 4G

Introduction

Random Access (RACH) is the MAC-layer handshake (TS 36.321) an LTE UE runs whenever it needs uplink synchronisation or a scheduling opportunity and has no other way to ask. It is the first uplink procedure in the connection lifecycle: before any scheduled uplink can flow, the eNB has to detect the UE, measure its round-trip delay, hand it a Timing Advance, and give it an initial grant.

It is not only for power-on. The same procedure runs for initial access from RRC_IDLE (RRCConnectionRequest), for RRC connection re-establishment after radio link or handover failure, for handover into a target cell, for regaining uplink sync when the timeAlignmentTimer has expired, for a Scheduling Request when no PUCCH SR resource is configured (or after SR failure), and for positioning when a fresh timing advance is required.

Because RACH is four distinct messages (Msg1–Msg4 in contention-based access, or two in the contention-free variant), a failure almost always lives in one specific step. That is exactly why a step-by-step troubleshooting habit pays off: if you know which message broke, you have already narrowed the root cause enormously.

Why Random Access is needed

Concretely, three things must happen before the UE can send any scheduled uplink. First, the eNB must detect the UE — it has no idea the UE exists until it hears a preamble. Second, it must synchronise the uplink: because the propagation delay is unknown, the eNB measures the round-trip and returns a Timing Advance so the UE's transmissions land inside the eNB's FFT window. Third, it must allocate initial uplink resources — a grant the UE uses to send its identity. RACH is the only uplink channel a UE can use before it holds any grant, so it bootstraps everything that follows.

When Random Access Is Used

Random Access (RACH) is not only for power-on. It is the generic MAC procedure (TS 36.321) the UE runs whenever it needs uplink synchronisation or a scheduling opportunity and has no other way to ask. Think of it as the single door a UE walks through any time it wants to talk uplink but the eNB either does not yet know it, or no longer has a valid timing relationship with it. The distinct triggers are:

• Initial access from RRC_IDLE — the UE has selected a cell, read MIB/SIB2, and now wants to set up a connection with RRCConnectionRequest. There is no prior uplink relationship at all, so contention-based access is the only option.
• RRC connection re-establishment — after radio link failure, handover failure, or integrity-check failure, the UE re-accesses a cell to recover with RRCConnectionReestablishmentRequest. It is racing the T311 timer, so speed matters.
• Handover — the UE accesses the target cell, usually with a dedicated (contention-free) preamble supplied inside the handover command's mobilityControlInfo. This keeps the mobility interruption short.
• Downlink data arrival when out of sync — the UE is RRC_CONNECTED but its uplink timing has expired (the timeAlignmentTimer lapsed), so before it can send HARQ feedback for a downlink transmission the eNB orders a RACH via a PDCCH order.
• Uplink data arrival when out of sync — the UE has UL data but no valid timing advance, or it is in sync but has no PUCCH resource configured for a Scheduling Request, so it falls back to RACH to regain sync and ask for a grant.
• Scheduling Request failure — the UE did have a dedicated PUCCH SR resource but exhausted dsr-TransMax without a grant; the SR procedure then falls back to a RACH.
• Positioning — when timing advance is needed for a UE-positioning method that requires it, the network can order RACH to obtain a fresh Timing Advance.

The Four-Step Contention-Based Procedure

This is the default when the UE has no dedicated resources. It is "contention-based" because many UEs share the same pool of preambles, and two of them can collide on the same choice. The four messages are conventionally called Msg1 through Msg4.

Msg1 — preamble on PRACH. The UE picks one of the up to 64 preambles configured for the cell and transmits it on a PRACH time/frequency resource. The set of 64 Zadoff-Chu sequences is split into groups signalled in SIB2; within the allowed group the UE chooses one randomly. From the time and frequency of the PRACH resource it used, the UE computes an RA-RNTI — a temporary identity that tells it which upcoming response is for its transmission (TS 36.321).

Msg2 — Random Access Response (RAR). The eNB, having detected the preamble, sends a MAC RAR on PDSCH scheduled by a PDCCH addressed to that RA-RNTI. The UE monitors for it inside the ra-ResponseWindowSize, a window that opens a few subframes after Msg1. The RAR carries an 11-bit Timing Advance command (so the UE corrects its uplink timing), a 20-bit uplink grant for Msg3, and a Temporary C-RNTI. It also echoes the detected preamble index (the RAPID, Random Access Preamble Identifier) so the UE confirms it was heard. One RAR MAC PDU can carry responses for several UEs, each keyed by its RAPID.

Msg3 — first scheduled uplink on PUSCH. Using the 20-bit grant, and scrambling with the Temporary C-RNTI, the UE transmits its identity — for initial access this is RRCConnectionRequest carrying the UE identity (S-TMSI or a 40-bit random value) on the CCCH. Msg3 is where contention actually bites: two UEs that chose the same preamble on the same resource both received the same RAR and now transmit on the same grant. Msg3 is HARQ-protected and can be retransmitted.

Msg4 — contention resolution. The eNB echoes back the identity it decoded, using the UE Contention Resolution Identity MAC CE (for a UE that already holds a C-RNTI, contention is instead resolved by a PDCCH addressed to that C-RNTI). The UE whose identity matches declares success and promotes its Temporary C-RNTI to C-RNTI; the loser fails contention resolution and retries.

Reading the four-step flow in the logs

Walk the trace message by message: preamble sent, RAR received in-window, Msg3 CRC at the eNB, Msg4 contention-resolution match. The first line that does not appear (or does not match) is the broken step.

Representative UE log (LTE initial access) — illustrative, values vary by vendor/build:

Preambles: Zadoff-Chu Sequences and Groups A/B

A preamble is not a random noise burst — it is a carefully chosen Zadoff-Chu (ZC) sequence. ZC sequences are "constant amplitude zero autocorrelation" (CAZAC): each has perfectly flat power across the band and, crucially, near-zero correlation with cyclically shifted versions of itself. That last property is what lets the eNB tell many simultaneous preambles apart and estimate the round-trip delay of each from where the correlation peak lands.

Each cell needs 64 preambles. They are generated from one or more root Zadoff-Chu sequences (identified by the rootSequenceIndex broadcast in SIB2) by applying cyclic shifts. The shift step is governed by zeroCorrelationZoneConfig (the NCS parameter): a larger zero-correlation zone tolerates larger propagation delay (bigger cells) but yields fewer shifts per root, so more root sequences are consumed to reach 64. If one root cannot produce 64 usable shifts, the UE simply rolls to the next consecutive root index — which is why rootSequenceIndex is a starting point, not a single value.

The 64 preambles are partitioned into up to two groups, A and B. The group tells the eNB, before it has decoded anything, roughly how big a Msg3 the UE intends to send and how good its coverage is. The UE selects Group B only when its Msg3 payload would exceed the messageSizeGroupA threshold and its estimated pathloss leaves enough power headroom (the eNB advertises the criteria via messagePowerOffsetGroupB); otherwise it uses Group A. This gives the eNB a hint to size the Msg2 grant appropriately.

Preamble Formats and PRACH Configuration

The Msg1 preamble's length in the time domain depends on the preamble format, chosen to match the cell size: larger cells need a longer cyclic prefix (CP) and guard time to tolerate the greater round-trip delay before the preamble smears into the next symbol. LTE defines formats 0–4 (sequence and CP lengths in TS 36.211; timing and cell-radius implications per TS 36.213). Formats 0–3 use the 1.08 MHz (6-RB) PRACH band in normal subframes; format 4 is the very short format living in the TDD special subframe's UpPTS.

The prach-ConfigIndex (broadcast in SIB2 within the prach-Config / PRACH-ConfigInfo structure) selects both the preamble format and which subframes and system-frame parity carry PRACH resources, from a lookup table in TS 36.211. Alongside it, prach-FreqOffset positions the 6-RB PRACH band in frequency, rootSequenceIndex seeds the ZC generation, and zeroCorrelationZoneConfig sets the cyclic-shift spacing. Together these tell every UE in the cell exactly when and where to send Msg1 and how the 64 preambles are generated — without any per-UE signalling. In TDD, PRACH occasions can additionally recur in frequency within a subframe, which is where the f_id term in the RA-RNTI earns its keep.

Contention Resolution: Two UEs, One Preamble

The word "contention" only means something when two UEs clash. Picture UE-A and UE-B starting random access at the same instant and happening to pick the same preamble on the same PRACH resource. The eNB detects one preamble energy at one delay and issues one RAR — which both UEs receive, because both computed the same RA-RNTI and both see their RAPID echoed. They now hold the same Temporary C-RNTI and the same grant, so they transmit Msg3 on the same PUSCH — a second collision.

The eNB typically decodes only the stronger Msg3 (capture effect). Msg4 then carries the UE Contention Resolution Identity MAC CE, which echoes back the first 48 bits of the CCCH SDU (the RRCConnectionRequest) it decoded. The UE whose transmitted identity matches those 48 bits declares success, stops the mac-ContentionResolutionTimer, and promotes Temporary C-RNTI to C-RNTI. The other UE sees a mismatch, declares failure when the mac-ContentionResolutionTimer expires, discards the Temporary C-RNTI, backs off by a random amount up to ra-ResponseWindowSize-related backoff, ramps power, picks a fresh random preamble, and retries.

There is a subtle second path. A UE that already has a valid C-RNTI (for example, doing RACH after losing uplink sync) does not use the identity MAC CE. It sends a C-RNTI MAC CE in Msg3, and contention is considered resolved the moment it receives a PDCCH addressed to its own C-RNTI — no 48-bit echo needed, because the C-RNTI is already unique in the cell.

Contention-Free (Non-Contention) Random Access

When the network already knows the UE and wants a fast, collision-proof access, it hands the UE a dedicated preamble in advance. Because no one else in the cell will be told to use that specific ra-PreambleIndex, there is no possibility of collision, and the procedure ends at Msg2 — no Msg3/Msg4 resolution is needed. This is the whole point: contention-free RACH trades a little signalling overhead (the eNB must pre-assign the preamble) for deterministic, minimum latency.

Two ways the eNB assigns the dedicated preamble:

• Handover command — the RRCConnectionReconfiguration carrying mobilityControlInfo includes a rach-ConfigDedicated with a reserved ra-PreambleIndex and ra-PRACH-MaskIndex, so the UE accesses the target cell deterministically the instant it arrives.
• PDCCH order — the eNB sends a DCI format 1A on the PDCCH (addressed to the UE's C-RNTI) carrying a specific preamble index, telling a RRC_CONNECTED but out-of-sync UE to launch RACH — typically to restore uplink timing before delivering pending downlink data, or to obtain a fresh Timing Advance for positioning.

Note a self-healing fallback: if the eNB assigned ra-PreambleIndex = 000000 in a PDCCH order, that instructs the UE to fall back to the ordinary contention-based procedure with a randomly chosen preamble. And if a dedicated attempt fails repeatedly (reaching preambleTransMax), the UE likewise falls back to contention-based access rather than giving up.

Power Ramping and Retransmission

The UE does not know the exact path loss to the cell, so it sets the first preamble power by open-loop estimation and then ramps it. If no valid RAR arrives within ra-ResponseWindowSize, or Msg4 does not resolve contention before mac-ContentionResolutionTimer expires, the UE increments its preamble transmission counter (PREAMBLE_TRANSMISSION_COUNTER), raises transmit power by powerRampingStep, waits a backoff drawn from the RAR's backoff indicator, picks a newly chosen random preamble, and tries again. This repeats until success or until the counter reaches preambleTransMax, at which point the MAC declares a random access problem to RRC (which may lead to re-establishment or cell reselection).

Here PL is the estimated downlink path loss (from the cell-specific reference-signal RSRP), preambleInitialReceivedTargetPower is the received level the eNB wants to see, and n is the current preamble attempt (TS 36.213). For a preamble chosen from Group B, an additional deltaPreambleMsg3-style offset and the group-B power offset apply. All of these higher-layer parameters — preambleInitialReceivedTargetPower, powerRampingStep, preambleTransMax, ra-ResponseWindowSize, mac-ContentionResolutionTimer, and the group-selection thresholds — are broadcast in the RACH-ConfigCommon of SIB2 for contention-based access, or delivered in rach-ConfigDedicated for a contention-free attempt.

Summary

The fastest way to root-cause an LTE random-access failure is to walk the messages in order and find the first one that broke. Msg1→Msg2→Msg3→Msg4: if the preamble never gets a RAR, suspect UE Tx power, RF/SINR, power ramping, or a PRACH config mismatch against SIB2. If the RAR is missing despite a detected preamble, look at RAR generation, PDCCH CCE allocation, RA-RNTI, and ra-ResponseWindowSize. If Msg3 fails, check CRC, the applied Timing Advance, and the HARQ retry budget. If Msg4 fails, check the UE Contention Resolution Identity match, mac-ContentionResolutionTimer, and downlink quality.

Keep the two flavours straight: contention-based access is "cold" (idle, re-establishment, UL data without SR) and needs the full four steps; contention-free access is network-initiated (handover, PDCCH order, positioning), uses a dedicated preamble, and finishes at Msg2. Isolating the broken step turns a vague "attach failure" into a specific, testable hypothesis — because each step points at a small, distinct set of causes.