Bandwidth Parts in 5G NR

• Bandwidth Parts are configurable segments of the total system bandwidth in 5G NR, defined by contiguous physical resource blocks with a single numerology.
• They enable dynamic switching and efficient resource allocation by allowing static, semi-persistent, or real-time configuration, balancing interference and power efficiency.
• Stochastic geometry models underpin the optimization of bandwidth partitioning, balancing SINR requirements with spatial reuse to maximize network performance.

A bandwidth part (BWP) is a technical concept for partitioning the total system bandwidth into narrower, contiguous or non-contiguous frequency segments, each potentially associated with distinct numerologies and configurable assignment to user equipment (UE). Originating in the context of decentralized wireless networks and now central to 5G New Radio (NR) architecture, the BWP paradigm enables fine-grained trade-offs between throughput, spatial reuse, interference, and power efficiency. In practical standards such as 5G NR, a BWP refers to a contiguous set of physical resource blocks (PRBs) within a carrier and has configuration, switching, and operational semantics defined by 3GPP. Stochastic geometry and probabilistic models underpin bandwidth-partitioning optimization in theoretical literature, while 5G NR standards leverage BWPs to support broad hardware heterogeneity, carrier aggregation, dynamic spectrum sharing, and energy-efficient operation.

1. Formal Definition and Configuration

In 5G NR, each carrier possesses a channel bandwidth (BW_carrier), divided into resource blocks (RBs), with each RB consisting of 12 subcarriers and a configurable subcarrier spacing denoted as Δf_p = 15 kHz·2^p for numerology index p. A BWP is defined as a contiguous sequence of N_{RB} RBs starting at index RB_start and associated with a single numerology (subcarrier spacing and cyclic prefix). The occupied bandwidth of a BWP is:

$$\mathrm{BW}_\mathrm{BWP} = N_{\mathrm{RB}} \times 12 \times \Delta f$$

where Δf denotes the subcarrier spacing in Hz (Lin et al., 2020, Jeon, 2017). A UE may be configured with up to four downlink and four uplink BWPs per serving cell but can only activate one per direction at any given time.

BWPs are configured initially during RRC (Radio Resource Control) procedures, which specify, for each BWP: its index, subcarrier spacing, cyclic prefix, startPRB, number of PRBs, and controlResourceSetList. Two configuration models exist for the initial BWP (#0): a minimal configuration for initial access or a fully UE-specific permanent BWP (Lin et al., 2020). RRC signaling can further configure up to four RRC BWPs per direction, with flexibility for static, semi-persistent, or dynamically switchable operation.

2. Bandwidth Partitioning and Theoretical Trade-offs

Bandwidth partitioning divides the total available spectrum W into N orthogonal sub-bands (or frequency slots), each of width W/N. In stochastic geometry models of decentralized networks (e.g., Poisson network models), each transmitter (TX) independently and randomly selects a sub-band for transmission over a fixed distance d to its paired receiver (RX). The system objective is, for a given data rate R and outage constraint ε, to select N maximizing the spatial density of simultaneous successful links, formally (0711.0277):

$$\lambda^T_\epsilon(N) = N \cdot \lambda_\epsilon(\beta(N), N_0 W/N)$$

where λ_ε is the maximum attempt density per sub-band that meets the outage constraint for SINR threshold β(N) and noise spectral density. The operating spectral efficiency on each sub-band is b = R/(W/N) [bps/Hz], related to required SINR threshold as β = 2^b - 1. The optimization of N involves a fundamental trade-off: increasing N reduces interference density but forces each link to operate at exponentially higher SINR due to shrinking bandwidth, raising the "exclusion area" for each link.

The optimal N* (number of sub-bands/BWPs) and corresponding spectral efficiency b* are derived from maximizing λ^T_ε with respect to system parameters (path loss exponent α, transmit power P, noise density N₀, total bandwidth W, per-link rate R) via fixed-point equations and stochastic geometry formulations. In the interference-limited regime (high SNR), N* is nearly constant for given W/R and α; in the power-limited regime, optimal N* depends on the received energy-per-bit (0711.0277).

3. BWP Assignment, Switching, and Operational Protocols

BWPs can be statically assigned, semi-persistently mapped, or dynamically switched. RRC-based switching involves a full RRC Reconfiguration message and latency on the order of 5–80 ms, after which the UE activates a new BWP, potentially also changing the numerology. Dynamically, the network can issue a BWP-indicator field in DCI (Downlink Control Information) formats, allowing per-transmission switching with sub-millisecond delay, governed by TBWPswitchDelay, which depends on the numerology and slot length (e.g., for Δf = 15 kHz, 4 slots ≈ 4 ms; Δf = 60 kHz, 16 slots ≈ 4 ms) (Lin et al., 2020). Timer-based switching uses an inactivity timer (T_inact) to revert the UE to the default BWP if no transmission opportunities are scheduled.

In infrastructureless device-to-device (D2D) models, bandwidth allocation can be adaptive: a wide carrier is split into M equal chunks of width w = W/M, and each user is assigned t contiguous chunks (BWP) reflecting its type and demand (Baccelli et al., 2021). These assignments can be contiguous (BWP-compliant in 5G NR) or randomly selected (non-contiguous, as in LTE).

4. Performance Analysis and Service Differentiation

The bandwidth part paradigm allows differentiation of service and resource allocation per user type and traffic profile. In theoretical adaptive BA models, users randomly select their type (i.e., the number of chunks occupied) with probability p_t (for type t), leading to bandwidth assignments and interference patterns governed by stochastic geometry (Baccelli et al., 2021).

For a type-k user, the success probability at a given SIR threshold θ is expressed as:

$$P_s^{(k)}(\theta)=\exp\left(-\lambda\,C\,\theta^\delta\sum_{i=1}^{M}p_i\sum_{t=0\vee(i+k-M)}^{\,i\wedge k}p^{(t)}_{k,i}\left(\frac{t}{k}\right)^\delta\right)$$

with δ = 2/α, C = πR^2Γ(1+δ)Γ(1−δ), and p^{(t)}_{k,i} the probability that a random interfering user of type i overlaps t chunks with the typical user.

Aggregate throughput exhibits two key properties:

• Egalitarian Per-Hz Service: The mean signal-to-mean-interference ratio is independent of user type; thus, each type experiences similar throughput per assigned Hz.
• Linear Aggregate Throughput Differentiation: The total throughput for a type-k user scales linearly with k, that is, T(k) ≈ k·T(1).

Higher traffic variability in per-type assignments increases Shannon throughput and reliability for given mean signal and interference, a direct consequence of Jensen's inequality applied to the Laplace exponents governing success metrics.

5. Practical 5G NR Use Cases and Constraints

Practical utilization of BWPs in 5G NR includes supporting UEs with limited RF/baseband capabilities, power savings, dynamic spectrum sharing, and carrier aggregation (Lin et al., 2020, Jeon, 2017). For example, a gNB may configure a 100 MHz carrier with 20 MHz BWPs for legacy or limited-bandwidth UEs, assigning larger BWPs on demand for high throughput applications or reverting to narrower BWPs to minimize RF power during inactivity (typically coordinated with DRX cycles).

Configurational and capability constraints govern the maximum/minimum BWP size (e.g., 1–275 RBs in Release 15), the number of BWPs per UE, and the supported numerologies per cell. UEs must support at least one DL and UL RRC-configured BWP; optional support for up to four BWPs and cross-numerology operation enables flexible resource partitioning in deployments.

BWP switching latencies are sufficiently low (20–200 μs in typical hardware estimates) to allow real-time adaptation even for uplink or ultra-reliable low-latency communications (URLLC), provided scheduler and baseband architectures accommodate dynamic retuning and resource mapping (Jeon, 2017).

6. Relation to Bandwidth Partitioning and Adaptive BA in Theoretical Models

The concept of partitioning bandwidth to improve spatial reuse and maximize concurrent transmission density has deep theoretical roots in stochastic geometry and information theory. The work by Andrews et al. formalizes the optimal partition, balancing exponential increases in SINR requirements with reductions in interference density, yielding system-specific fixed-point equations for optimal spectral efficiency and partitioning (0711.0277).

Adaptive BA models, such as those in (Baccelli et al., 2021), extend the BWP paradigm into infrastructureless networks with probabilistic user demand, enabling both service differentiation and performance quantification. The restriction to contiguous chunks (as mandated by BWP design) yields slightly lower performance than random chunk allocation but preserves critical per-Hz and aggregate-rate properties.

Comparative analysis with LTE carrier aggregation highlights the lower control overhead, faster switching, and reduced RF complexity enabled by BWPs, especially in wideband or heterogeneous hardware environments (Jeon, 2017).

---

References:

• Andrews, J.G., et al., "Bandwidth Partitioning in Decentralized Wireless Networks" (0711.0277)
• Dahlman, E., et al., "A Primer on Bandwidth Parts in 5G New Radio" (Lin et al., 2020)
• Jeon, J., et al., "NR Wide Bandwidth Operations" (Jeon, 2017)
• Baccelli, F., Kalamkar, S., "Bandwidth Allocation and Service Differentiation in D2D Wireless Networks" (Baccelli et al., 2021)