Resource Grid: RE, RB & BWP in 5G NR
Resource elements, resource blocks, the common resource grid, and Bandwidth Parts.
Everything the 5G physical layer transmits is painted onto a two-dimensional canvas: frequency (subcarriers) across, time (OFDM symbols) along. The smallest cell of that canvas is a resource element; twelve stacked subcarriers make a resource block; a shared origin called Point A pins every block to an absolute frequency; and a UE only ever operates inside a slice of the whole carrier at a time — its active bandwidth part. Get these units, their reference point, and the three flavours of RB numbering straight, and the rest of NR scheduling falls into place. This is TS 38.211 territory, with the BWP and RIV procedures in TS 38.213 and the carrier/BWP IEs in TS 38.331.
Introduction
The resource grid is the coordinate system for the entire NR physical layer. Every transmitted signal — PDSCH data, PDCCH control, DMRS and CSI-RS reference signals, the SSB — is nothing more than energy placed on specific cells of a time-by-frequency lattice. Before a UE can be scheduled anything, it and the gNB must agree exactly which cells of that lattice a grant refers to, and at what absolute frequency each one sits. The resource grid, its reference point, and its numbering conventions are that shared language.
Three ideas do most of the work. First, the atomic units: the resource element (one subcarrier × one symbol) and the resource block (12 subcarriers). Second, the anchoring: Point A gives every block an absolute home on the frequency axis, and the Common Resource Block grid is the ruler laid from it. Third, the UE's working window: the bandwidth part, a contiguous slice of the carrier with its own numerology that the UE actually receives, so it need not process a 400 MHz carrier all at once.
This page builds from the RE and RB up through the three kinds of RB numbering (CRB, PRB, VRB), the Point A anchoring with offsetToPointA and k-SSB, the per-numerology common grid, bandwidth parts and their RIV encoding, the VRB-to-PRB mapping, and finally the per-antenna-port stack of grids that MIMO needs. Definitions come from TS 38.211; the BWP RIV and switching procedures from TS 38.213; and the IEs (SCS-SpecificCarrier, BWP, offsetToPointA) from TS 38.331.
On this page
Why the resource grid is needed
In plain words: think of the whole carrier as a huge spreadsheet — columns are moments in time (OFDM symbols), rows are frequencies (subcarriers). Every cell holds one tiny piece of signal. To tell a phone "your data is in these cells," everyone must agree where cell (0,0) is and how the rows and columns are numbered. Point A is the top-left corner everyone shares; the bandwidth part is the block of rows a given phone is asked to read; and CRB/PRB/VRB are just three different ways of numbering the rows for three different purposes.
The scheduler has to hand out fractions of a very large, shared resource to many UEs at once, second by second, without collisions and without ambiguity. A two-dimensional grid makes that tractable: it turns "who transmits what, where and when" into a set of rectangles on a lattice. Because OFDM already gives orthogonal subcarriers in frequency and discrete symbols in time, the grid is the natural bookkeeping for the air interface — it is what a DCI grant actually points at.
But a grid alone is not enough. UEs differ in bandwidth, use different numerologies, and each look at only part of the carrier. So NR needs (a) a single absolute reference — Point A — so that a block means the same frequency to every UE regardless of its own bandwidth or SCS, and (b) a per-UE window — the BWP — so a device with a narrow receiver can still be told precisely where to tune. The rest of this page is the machinery that makes those two things exact.
Resource elements and resource blocks
Start at the atom. A RE (resource element) is one subcarrier in frequency by one OFDM symbol in time — the smallest addressable piece of the grid, carrying exactly one complex modulation symbol on one antenna port. Stack twelve consecutive subcarriers and you get a RB (resource block), the unit the scheduler thinks in.
A RE = 1 subcarrier × 1 OFDM symbol, indexed by the pair (k, l) where k is the subcarrier and l the symbol. A RB = 12 consecutive subcarriers — in the frequency domain only. An NR resource block has no fixed time duration, unlike LTE.
The RE is the natural quantum for placing data, reference signals and control. Grouping into 12-subcarrier RBs keeps allocation and signalling manageable — the scheduler grants RBs, not individual subcarriers, and DMRS/CSI-RS patterns repeat neatly on a 12-subcarrier period.
A physical channel such as PDSCH is a rectangle of REs: some number of RBs in frequency × some number of symbols in time, with certain REs punctured for DMRS, PT-RS and other reference signals so the receiver can estimate the channel.
The RE carries one complex modulation symbol whose bit count depends on the modulation order: 2 bits for QPSK, 4 for 16-QAM, 6 for 64-QAM, 8 for 256-QAM (and 10 for the Rel-17 1024-QAM downlink). Each RE belongs to a specific antenna port, which is why two MIMO layers can occupy the same (k, l) cell on different ports. The index pair runs k over subcarriers within the grid and l over the OFDM symbols of the slot (0…13 under normal CP).
NR vs LTE gotcha: in NR a resource block is 12 subcarriers, full stop — a purely frequency-domain concept. There is no LTE-style "one RB = 12 subcarriers × 7 symbols" pairing. The absolute width of a RB in Hz therefore depends entirely on the subcarrier spacing (the numerology μ).
Because a RB is a fixed count of subcarriers rather than a fixed slice of spectrum, its width in hertz scales with the subcarrier spacing (SCS). The RB is 12 × SCS, and NR defines the numerologies as μ = 0…4 with SCS = 2μ × 15 kHz. Data channels use μ = 0…3 (15/30/60/120 kHz); the 240 kHz spacing is reserved for the SSB in FR2. Release 17 adds 480 and 960 kHz for FR2-2.
Numerology μ | SCS | RB bandwidth (12 × SCS) | Typical use |
|---|---|---|---|
| 0 | 15 kHz | 180 kHz | FR1 data (default), coexistence with LTE |
| 1 | 30 kHz | 360 kHz | FR1 data (most common mid-band) |
| 2 | 60 kHz | 720 kHz | FR1/FR2 data, URLLC |
| 3 | 120 kHz | 1.44 MHz | FR2 data |
| 4 | 240 kHz | 2.88 MHz | FR2 SSB only (not for data) |
| 5 / 6 | 480 / 960 kHz | 5.76 / 11.52 MHz | FR2-2 (Release 17), >52.6 GHz |
One more count worth memorising: a carrier can span up to 275 RBs per numerology, and a single carrier is at most 3300 subcarriers wide (275 × 12). That 275 ceiling reappears everywhere — in the RIV formula for BWPs and in the maximum channel bandwidth per numerology.
LTE ↔ NR: in LTE a resource block was a two-dimensional tile — 12 subcarriers × one 0.5 ms slot of 7 symbols — and the spacing was always 15 kHz, so an RB was always 180 kHz. NR redefines the RB as purely 12 subcarriers with no time dimension, and lets the spacing scale, so RB width ranges from 180 kHz (15 kHz) to 2.88 MHz (240 kHz). NR also adds the Point A / CRB absolute grid and the BWP concept, neither of which exists in LTE, where the UE always spanned the whole (much narrower) carrier.
Three kinds of RB numbering: CRB, PRB, VRB
The same physical resource block can be named three different ways depending on what you are describing. Confusing them is the classic NR trap, so pin each one to the question it answers.
| Term | Numbered from | Answers | Where it lives |
|---|---|---|---|
CRB (Common Resource Block) | Point A (CRB 0) upward, per numerology | "Where on the carrier, absolutely?" | The common grid; carrier & BWP placement |
PRB (Physical Resource Block) | 0 at the start of a given BWP | "Where inside this BWP?" | Scheduling grants, DCI frequency allocation |
VRB (Virtual Resource Block) | 0 at the start of the BWP (logical) | "Which logical block did the DCI assign?" | DCI allocation before VRB-to-PRB mapping |
CRBs are the absolute ruler laid from Point A. PRBs are the local ruler inside a BWP, starting at 0. VRBs are logical blocks a DCI allocates, later mapped to physical PRBs.
CRBs give every UE, numerology and BWP a common absolute reference. PRBs keep the scheduling grant compact — a DCI need only address blocks within the active BWP. VRBs add a layer of indirection so the network can interleave a contiguous allocation across the band for frequency diversity.
They relate by simple offsets: PRB = CRB − NBWP,start, where NBWP,start is the BWP's first CRB. A DCI allocates VRBs; a mapping rule (below) turns those into PRBs; the CRB layer keeps the whole thing anchored to real hertz.
In practice a scheduler grants a set of VRBs in the DCI, the UE converts VRB→PRB by the configured mapping, adds the BWP start to get the CRB, and finally uses Point A and the numerology to know the exact subcarrier frequencies. Four names, one chain, one physical block. The DCI does not spell out a start and length as two integers; it packs them into a single Resource Indication Value (RIV) under resource allocation type 1 — the very same encoding trick used for the BWP's locationAndBandwidth, covered below.
Point A, offsetToPointA and k-SSB
To place any RB on an absolute frequency axis, everyone needs a shared origin. That origin is Point A, and the ruler laid out from it is the Common Resource Block grid. Point A is not necessarily inside the carrier or the SSB — it is simply the agreed zero of the frequency axis.
Point A is the common reference point for all resource grids — it coincides with the centre of subcarrier 0 of Common Resource Block 0 for subcarrier spacing configuration μ = 0 (15 kHz). Every CRB grid, at every numerology, is anchored so that its subcarrier 0 aligns to Point A.
Different UEs, numerologies and bandwidth parts must agree on where a given RB sits. Anchoring everything to Point A means an offset in CRBs unambiguously identifies a frequency location, regardless of a UE's own bandwidth or the SCS it happens to use.
Point A is derived one of two ways: from offsetToPointA (for the SpCell during initial access, given relative to the SSB) or from absoluteFrequencyPointA (an explicit ARFCN, for other carriers). Everything else — carriers, BWPs — is then expressed as CRB offsets from it.
Two closely related offsets connect the SSB the UE just found to this common grid, and they are a frequent source of confusion:
offsetToPointA— the distance, counted in RBs of 15 kHz for FR1 (60 kHz for FR2), from Point A up to the lowest subcarrier of the common RB that overlaps the SSB. It is signalled in SIB1 insidefrequencyInfoDL. Its range is 0–2199 RBs. This is coarse, RB-granular tuning: it tells the UE where the SSB block sits on the common ruler.k-SSB(a.k.a.ssb-SubcarrierOffset) — the fine, subcarrier-granular gap between subcarrier 0 of that common RB and subcarrier 0 of the SSB itself. Because the SSB has its own 20-RB grid that need not align to the common grid,k-SSBmops up the leftover subcarriers (0–23 in 15 kHz units for FR1). It is carried in the MIB (with an extra MSB signalled separately for FR1 viaPBCHpayload).
Coarse plus fine: think of offsetToPointA as the whole-RB part and k-SSB as the sub-RB remainder. Together they let a UE that has only decoded the MIB compute exactly where Point A lies relative to the SSB it is camped on, and thus reconstruct the entire common grid before it has any dedicated configuration.
The per-numerology common grid and SCS-SpecificCarrier
Point A defines CRB 0, but a carrier can be received with several numerologies at once, and each numerology gets its own common resource grid. NR describes each of those grids with a SCS-SpecificCarrier element.
A SCS-SpecificCarrier describes the carrier for one subcarrier spacing: its subcarrierSpacing, its offsetToCarrier (the number of RBs, at that SCS, from Point A up to the carrier's lowest usable subcarrier), and its carrierBandwidth (the carrier width in RBs, up to 275).
Because CRB numbering is defined per numerology, the network must state, for each SCS it uses, where the carrier begins on that SCS's ruler and how wide it is. One physical carrier therefore carries a small list of these elements — one per active numerology.
The list lives in frequencyInfoDL / frequencyInfoUL (broadcast in SIB1) and in the dedicated ServingCellConfigCommon. Each entry says: "at 30 kHz, the carrier starts offsetToCarrier RBs above Point A and is carrierBandwidth RBs wide."
The field ranges make the picture concrete: offsetToCarrier is an INTEGER 0–2199 (RBs at this SCS above Point A), and carrierBandwidth is 1–275 RBs. Because the same physical spectrum is covered by fewer, wider CRBs at higher SCS, the numeric offsetToCarrier and carrierBandwidth differ per numerology even though they describe the same carrier edges — which is exactly why one SCS-SpecificCarrier entry is needed per active spacing.
offsetToPointA vs offsetToCarrier: offsetToPointA locates the SSB relative to Point A (always in 15/60 kHz units). offsetToCarrier locates the carrier edge relative to Point A, expressed in RBs of the specific numerology it applies to. Different reference objects, different units — do not swap them.
The CRB index of a given subcarrier follows directly from the numerology. Once Point A and offsetToCarrier are known, the UE builds the grid: CRB 0 sits at Point A, CRB 1 is the next 12 subcarriers up, and so on, at whatever SCS this grid uses. A wider SCS means fewer, wider CRBs cover the same spectrum, which is exactly why each numerology needs its own count.
Bandwidth Parts
A carrier can be very wide — up to 400 MHz in FR2. Forcing a UE to receive across all of it, all the time, would drain the battery for no benefit. The bandwidth part lets the UE work in just a slice of the carrier, with its own numerology, and change that slice on demand.
A BWP (bandwidth part) is a contiguous set of Common Resource Blocks on a carrier, with its own subcarrierSpacing, its own cyclicPrefix, and a location given by a start CRB and a length in RBs. Those two numbers are packed into one field, locationAndBandwidth.
Two big wins: bandwidth adaptation (a narrow BWP when traffic is light, a wide one when it is heavy) and power saving (a UE monitoring only a narrow BWP runs a cheaper receiver). It also lets different services run different numerologies on one carrier.
Up to 4 BWPs can be configured per direction, but only 1 DL BWP and 1 UL BWP are active at a time per serving cell. At initial access the UE uses the initialDownlinkBWP / initialUplinkBWP derived from SIB1, before dedicated BWPs arrive over RRC.
The clever part is locationAndBandwidth. Rather than sending a start and a length separately, NR encodes both into a single Resource Indication Value (RIV), the same trick used for resource allocation type 1. With N = 275, a starting RB RBstart and a length L (1 ≤ L ≤ 275 − RBstart):
The UE inverts this to recover RBstart and L. The BWP's start is measured relative to CRB 0 (Point A) for the initial BWP context, but note the RIV itself is computed with respect to the carrier / a size of 275; the resulting start CRB is where PRB 0 of this BWP sits. From there, PRBs run 0…(L−1) inside the BWP. The single packed value fits a fixed-width DCI/RRC field of ⌈log2(N(N+1)/2)⌉ bits, which is the whole point of the RIV encoding — one compact field carries both numbers unambiguously.
| Concept | Value / rule |
|---|---|
| Configured BWPs per direction | Up to 4 (plus the initial BWP) |
| Active at any time | 1 DL + 1 UL per serving cell (for paired/TDD; SUL adds its own) |
| Initial BWP | initialDownlinkBWP / initialUplinkBWP, from SIB1, used before dedicated config |
| Default BWP | defaultDownlinkBWP-Id; UE falls back here on inactivity (defaults to the initial DL BWP if not set) |
| Each BWP carries | Its own subcarrierSpacing, cyclicPrefix, and locationAndBandwidth (RIV) |
| Switch triggers | DCI (with a bandwidthPartIndicator), bwp-InactivityTimer expiry, RRC reconfig, or RACH |
BWP switching: the network can move a UE to another configured BWP with a scheduling DCI (format 0_1/1_1) carrying a bandwidthPartIndicator, or it can drop the UE to the default BWP automatically when the bwp-InactivityTimer expires after a spell of no scheduling — a clean power-saving reflex. Only one DL and one UL BWP are ever active, so a switch deactivates the previous one.
The distinction between initial, active and default BWP matters in interviews: the initial BWP is what you bootstrap on from SIB1; the active BWP is whichever one is live right now; the default BWP is where the inactivity timer sends you to save power. Early in a connection all three can be the same BWP.
VRB-to-PRB mapping
The DCI allocates virtual resource blocks; the physical layer then decides which physical RBs actually carry them. That indirection is the VRB-to-PRB mapping, and it comes in two modes.
Non-interleaved mapping is the identity: VRB n → PRB n within the BWP. Interleaved mapping shuffles VRBs across the BWP in bundles, so a contiguous VRB allocation lands on PRBs scattered across the band.
Interleaving buys frequency diversity: a small allocation is spread over the whole BWP so that a deep fade in one part of the band cannot wipe out the whole transmission. Non-interleaved keeps blocks contiguous, which suits frequency-selective scheduling and beam/precoder localisation.
A 1-bit field in the DCI (VRB-to-PRB mapping) selects the mode when both are allowed. Interleaving works on RB bundles of size L (2 or 4, from vrb-ToPRB-Interleaver) using a block interleaver defined in TS 38.211.
For PDSCH the bundle size L is 2 or 4 RBs; the interleaver writes bundles by rows and reads them by columns, producing the scattered PRB pattern. CORESET#0-scheduled PDSCH and some common allocations use fixed interleaving parameters (bundle size 2) so a UE without dedicated config can still follow the mapping. PUSCH, by contrast, always uses non-interleaved mapping in current releases.
One-line summary: non-interleaved = "what you allocate is what you get" (VRB n = PRB n); interleaved = "spread it out for diversity." The DCI bit picks the mode; the bundle size sets the granularity of the shuffle.
Antenna-port grids and the whole picture
One subtlety hides behind the flat two-dimensional picture: there is not a single grid but a stack of them. TS 38.211 defines one resource grid per antenna port, per numerology, per transmission direction. A signal transmitted on antenna port p at subcarrier spacing configuration μ has its own (k, l) grid.
An antenna-port grid is a complete time-frequency grid of REs associated with one antenna port p and one numerology μ. DMRS ports, CSI-RS ports and data layers each live on their own logical grid.
MIMO and beamforming transmit different (precoded) content on different ports simultaneously on the same REs. Modelling one grid per port keeps the mapping from layers/ports to physical antennas clean, and lets the receiver estimate each port's channel from its DMRS.
Antenna ports are grouped by signal type (e.g. 1000-series for PDSCH DMRS, 3000-series for CSI-RS). The physical-layer mapping places each port's REs; precoding then combines ports onto the physical antenna elements.
The port-number ranges are themselves a naming convention in TS 38.211: the 0-series is PBCH/PSS/SSS, the 1000-series is PDSCH DMRS, the 2000-series is PDCCH DMRS, the 3000-series is CSI-RS, and on the uplink the 0-series covers PUSCH DMRS. Because each port has its own grid, a rank-4 PDSCH transmission is four DMRS ports (1000–1003) sharing the same REs but distinguished by orthogonal DMRS — which is exactly what lets the receiver separate the four spatial layers.
The figure below ties the whole hierarchy together: REs form RBs, RBs are numbered as CRBs from Point A across the carrier, a bandwidth part is a contiguous window of RBs the UE actually operates in, and each antenna port sees its own copy of the grid.
Spec anchors: the grid, RE/RB definitions and antenna-port grids are in TS 38.211; the numerologies and 275-RB limit likewise. BWP RIV encoding and switching live in TS 38.213; the IEs (SCS-SpecificCarrier, BWP, locationAndBandwidth, offsetToPointA) are in TS 38.331.
⚠ Common pitfalls / gotchas
- Swapping
offsetToPointA(locates the SSB, always in 15/60 kHz RB units) withoffsetToCarrier(locates the carrier edge, in RBs of the specific numerology) — different reference objects and different units. - Treating an NR RB as a 2-D tile like LTE. It is 12 subcarriers only; no time dimension, and its Hz width changes with SCS.
- Reading a DCI frequency allocation as CRB or PRB when it is actually VRB — you must apply the VRB-to-PRB mapping (and add
NBWP,start) before it means an absolute frequency. - Forgetting that only one DL and one UL BWP are active per cell, so a
bandwidthPartIndicatorswitch silently deactivates the previous BWP — anything scheduled on the old BWP is gone. - Assuming k-SSB is in the same units as offsetToPointA — k-SSB is subcarrier-granular (the sub-RB remainder), not RB-granular.
Summary
The NR resource grid is one coordinate system built from small, exact pieces. The atom is the RE (one subcarrier × one symbol, one modulation symbol per antenna port); twelve subcarriers make an RB, a frequency-only unit whose width is 12 × SCS and which scales from 180 kHz to 2.88 MHz across the numerologies, with a carrier capped at 275 RBs. Absolute placement comes from Point A, the shared zero from which the per-numerology Common Resource Block grid is measured; offsetToPointA (coarse, RB-granular) and k-SSB (fine, subcarrier-granular) tie the SSB to that grid, and SCS-SpecificCarrier states each numerology's carrier start and width.
On top of the absolute grid, each UE works inside a BWP — a contiguous CRB window with its own SCS and CP, its start and length packed into one locationAndBandwidth RIV, up to 4 configured per direction but only one active at a time. Three RB numberings answer three questions — CRB (absolute from Point A), PRB (local to the BWP), VRB (logical, in the DCI) — linked by PRB = CRB − NBWP,start and the VRB-to-PRB mapping (non-interleaved for locality, interleaved for frequency diversity). Finally, MIMO stacks one grid per antenna port on the same REs. Keep the units, the reference point, and the three numberings straight and NR scheduling reads cleanly.
Quick Q&A
Q. Define RE and RB in NR — and the LTE difference.
A. A RE is one subcarrier × one OFDM symbol. A RB is 12 consecutive subcarriers — a frequency-domain-only concept in NR (no fixed time duration), unlike LTE where a RB was 12 subcarriers × a slot of symbols. A RB's width in Hz is 12 × SCS, so it scales with numerology.
Q. Distinguish CRB, PRB and VRB.
A. CRB is numbered from Point A (absolute, per numerology); PRB is numbered from 0 at the BWP start; VRB is the logical block a DCI allocates, mapped to PRBs by the VRB-to-PRB rule. Relation: PRB = CRB − NBWP,start.
Q. What is Point A, and how do offsetToPointA and k-SSB differ?
A. Point A is the common reference — subcarrier 0 of CRB 0 at 15 kHz. offsetToPointA is the coarse, RB-granular distance from Point A to the common RB overlapping the SSB (15/60 kHz units, in SIB1). k-SSB is the fine, subcarrier-granular offset from that common RB's subcarrier 0 to the SSB's subcarrier 0 (in the MIB).
Q. What does SCS-SpecificCarrier carry, and what is offsetToCarrier?
A. It describes the carrier for one numerology: subcarrierSpacing, carrierBandwidth (RBs, up to 275) and offsetToCarrier — the number of RBs at that SCS from Point A to the carrier's lowest usable RB. There is one such element per active numerology.
Q. How many BWPs, and how is location encoded?
A. Up to 4 per direction, 1 DL + 1 UL active at a time. Start and length are packed into locationAndBandwidth as an RIV with N = 275. Switching is by DCI bandwidthPartIndicator or bwp-InactivityTimer expiry to the default BWP; the initialDownlinkBWP from SIB1 is used before dedicated config.
Q. When would you use interleaved VRB-to-PRB mapping?
A. When you want frequency diversity — a small or common allocation spread across the BWP in bundles (size 2 or 4) so a deep fade cannot kill it. Non-interleaved (VRB n = PRB n) suits frequency-selective scheduling and localised precoding; a DCI bit selects the mode for PDSCH, while PUSCH stays non-interleaved.
Where this fits
You now have the frequency map of the grid and its reference points. Next, see how the grid is divided in time, and how control channels reserve their own corner of it.