5G Interview Questions: What Gets Asked and How to Answer It
A structured 5G interview guide: the questions that actually come up across radio, RAN, core and QoS, worked troubleshooting and system-design scenarios, and a framework for answering well.
Most 5G interviews aren't testing whether you've memorised specifications. They're testing whether you can move between three levels — what a thing is, how it fits the architecture, and what trade-off it represents — without getting stuck at the first one.
Candidates who fail usually fail in a specific way: they give a textbook definition, stop, and wait. Candidates who do well answer the question, connect it to the surrounding architecture, and name the trade-off unprompted.
This guide is organised by topic rather than as a numbered list, so you can work through the area you're weakest in. Each section links to the deeper article if you need to build the underlying understanding first.
The answering framework
Before the questions, the structure that makes answers land.
Level 1 — define it. One or two sentences. Precise, no hedging.
Level 2 — place it. Which layer, which node, which interface, what it talks to.
Level 3 — trade it off. What it costs, when it doesn't apply, what you'd give up.
Most candidates deliver Level 1 and stop. Adding Level 2 marks you as someone who has seen a real network. Adding Level 3 marks you as someone who has made design decisions.
Two habits worth building. Don't dive into message-level detail unprompted — it reads as narrow rather than deep. And say "I don't know" cleanly when you don't, then say how you'd find out. Interviewers are testing calibration as much as knowledge, and confident wrong answers score far worse than honest gaps.
Radio and air interface
Deeper: What is 5G NR? · Beamforming · Massive MIMO
What is 5G NR?
The 3GPP radio access technology for 5G — the air interface between UE and gNB, covering PHY, MAC, RLC, PDCP and SDAP.
The Level 2 addition: NR is a framework rather than a fixed configuration. Where LTE fixed subcarrier spacing, slot duration, and carrier bandwidth, NR parameterises all three. That single design choice explains most of what follows.
What is numerology?
Subcarrier spacing scales as Δf = 2^µ × 15 kHz, giving 15, 30, 60, 120 and 240 kHz. A slot is always 14 symbols, so doubling the spacing halves the slot duration.
The trade-off is where the marks are: wider spacing gives shorter slots and better Doppler and phase-noise tolerance, but a proportionally shorter cyclic prefix, so less delay-spread tolerance. That's why 15 kHz survives for wide-area low band, 30 kHz dominates mid-band, and 120 kHz is used at mmWave.
How does NR achieve low latency?
Three mechanisms, and naming all three is the strong answer: scalable numerology shortens slots; mini-slots let transmission start at any symbol rather than waiting for a slot boundary; flexible HARQ timing via K1 and K2 allows self-contained slots where data and acknowledgement occupy the same slot.
What is beamforming, and what are the architectures?
Beamforming applies phase offsets across antenna elements so wavefronts add constructively in one direction.
The architectural question is really about RF chain count: analog uses one chain for the whole array (cheap, one beam at a time), digital uses one per element (maximum flexibility, prohibitive at scale), hybrid uses one per subarray (the practical compromise). FR1 typically goes digital; FR2 goes hybrid because 256 full chains at mmWave isn't feasible.
Why does massive MIMO increase capacity?
The answer that separates candidates: not beamforming — MU-MIMO. Serving multiple users on the same time-frequency resources, separated spatially.
That makes it a cell-capacity technology whose gains scale with load. An empty cell sees no benefit because there's nobody to pair. Follow up with why TDD matters: channel reciprocity means the gNB infers the downlink channel from uplink SRS, so feedback overhead doesn't scale with array size.
RAN, scheduling and HARQ
Deeper: The gNB explained · Scheduling algorithms · HARQ · RRC states
What does the gNB do?
Terminates NR toward the UE, connects to the 5G Core over NG. Handles scheduling, link adaptation, HARQ, mobility, RRC connection management, QoS enforcement via SDAP, and PDCP security.
Level 2: it can be split — CU hosting RRC/SDAP/PDCP, DU hosting RLC/MAC/PHY, with the cut between PDCP and RLC because that's where hard real-time ends. The CU can split further into CU-CP and CU-UP over E1.
What does the MAC scheduler do?
Decides, every slot, which UEs to serve, how many resource blocks each gets, and which MCS to use — using channel quality, buffer status, QoS parameters, packet delay, and HARQ state.
The point worth making unprompted: 3GPP doesn't specify the algorithm. It specifies parameters and expected outcomes. Scheduling is where vendors differentiate, and two conformant gNBs can behave quite differently under load.
Round Robin vs Proportional Fair
Round Robin serves each UE in turn with equal resources, ignoring the channel. Simple, starvation-free, and poor — because equal resources produce very unequal throughput depending on channel conditions.
Proportional Fair schedules on instantaneous achievable rate divided by average throughput. The elegance is that it exploits multi-user diversity: fading is independent across UEs, so someone is always at a peak relative to their own average, and serving them is both efficient and fair.
PF's gap is that it knows nothing about QoS. Production schedulers layer on top: retransmissions first, then GBR commitments, then delay-critical traffic, then weighted PF.
What is HARQ, and what does soft combining buy you?
FEC plus retransmission, where failed transmissions are retained in a soft buffer and combined rather than discarded.
The strong follow-up: HARQ is what makes aggressive link adaptation viable. The scheduler can target ~10% first-transmission BLER instead of 0.1%, because failures are cheap to recover — which means higher MCS and higher average throughput. HARQ isn't a safety net; it's a throughput enabler.
Chase combining vs incremental redundancy
Chase combining retransmits an identical copy — you're adding energy only. Incremental redundancy sends different coded bits from the same mother code, genuinely lowering the effective code rate with each attempt.
NR uses IR with four redundancy versions (RV0–RV3), RV0 being self-decodable. Chase combining is effectively the special case where the scheduler resends RV0.
Why multiple HARQ processes?
Because stop-and-wait would idle the link for a full round trip. NR supports up to 16 parallel processes per carrier per direction — double LTE's eight, because NR's shorter slots fit more transmissions inside one round trip.
Worth adding: HARQ delivers out of order, which is exactly why RLC AM sits above it doing reordering.
Why was RRC_INACTIVE introduced?
Because LTE's two states forced a bad trade-off on bursty traffic. Stay CONNECTED and burn battery on an idle connection; drop to IDLE and pay full setup — around 50–100 ms — for every small message.
INACTIVE retains the UE context at the anchor gNB while the UE behaves like it's idle. Resume costs ~10–20 ms instead. The detail that shows depth: the core network never knows the UE went inactive — the NG connection stays up, so resuming skips the entire core signalling path.
Core network and sessions
Deeper: 5G Core architecture · AMF vs SMF vs UPF · PDU Sessions
Explain the 5G Core architecture
Service-based: network functions expose HTTP/2 APIs and discover each other through the NRF, rather than communicating over fixed point-to-point reference points.
Level 3 — the nuance that impresses: not everything is service-based. N1 (NAS), N2 (NGAP), N3 and N9 (GTP-U), N4 (PFCP) and N6 stay purpose-built, because HTTP/2 and JSON are fine for session-setup-rate signalling and entirely wrong for forwarding packets at line rate.
AMF vs SMF vs UPF
AMF manages the device: registration, mobility, paging, NAS security. SMF manages the session: lifecycle, IP allocation, UPF selection, policy translation. UPF carries the packets: forwarding, QoS enforcement, usage reporting.
The detail interviewers probe: the AMF does not do session management, but session-management NAS still passes through it. It routes the SM container to the SMF without interpreting it — which is why one AMF can serve a UE whose sessions terminate at different SMFs in different slices.
Add the reason for the split: they scale on completely different axes — devices, sessions, traffic volume.
What is a PDU Session, and how is it established?
Logical connectivity between UE and a Data Network, established and controlled by the 5G Core, belonging to exactly one slice with one PDU type.
The flow in five beats: UE sends the request as NAS → AMF selects an SMF via NRF filtered by S-NSSAI and DNN → SMF retrieves policy, allocates the address, selects a UPF → SMF programs the UPF over N4 with PDR/FAR/QER/URR → Accept returns with address and QoS rules.
The line worth delivering: neither the AMF nor the SMF ever sees a user packet.
Control plane vs user plane
Control plane carries signalling — registration, session management, mobility, policy. User plane carries subscriber traffic.
Level 3: separation is what allows independent scaling and edge UPF placement. Without CUPS there's no low-latency edge breakout, because the anchor would always be central.
What is network slicing?
Multiple logical networks over shared infrastructure, identified by S-NSSAI = SST (8-bit service type) + optional SD (24-bit differentiator).
The two points that separate a good answer: requested does not mean allowed — the network decides per registration area based on subscription, policy and availability. And slice selection is not resource reservation — selection establishes context; isolation requires separate RAN, transport and Core resource management.
Also worth knowing: a PDU Session belongs to exactly one slice instance, so a UE needing two slices needs two sessions.
QoS
Deeper: QoS Flows
What is a QoS Flow, and how does it differ from a PDU Session?
The finest granularity of QoS differentiation inside a PDU Session. The session provides connectivity; flows provide differentiated treatment within it.
QFI vs 5QI — and what about ARP?
QFI identifies which flow a packet belongs to — 6 bits, scoped to one PDU Session, no inherent meaning. 5QI points to a set of QoS characteristics — resource type, priority, packet delay budget, packet error rate — scoped network-wide.
Adding ARP unprompted is a strong move. It governs admission and preemption, which is separate from treatment: a flow can have excellent 5QI characteristics and still be refused admission because its ARP is low.
The trap to avoid: 5QI is not a priority number. 5QI 3 has worse priority than 5QI 5 but a tighter delay budget and looser error target. They're specified for different failure modes.
GBR vs Non-GBR
Non-GBR shares capacity under Session-AMBR and UE-AMBR with no rate guarantee. GBR reserves resources for a guaranteed flow bit rate and is admission-controlled.
Delay-critical GBR is the one to mention: it adds Maximum Data Burst Volume and applies a different delay semantic — packets exceeding the PDB count as errors, not merely late. For a 5 ms control loop, data at 8 ms is useless, so the specification treats it as lost.
How does QoS reach the radio scheduler?
The chain: PCF issues PCC rules → SMF translates into QoS profiles and rules → profile goes to the RAN over N2, packet detection rules to the UPF over N4 → UPF marks downlink packets with QFI in the GTP-U header → gNB looks up the profile → SDAP maps the flow onto a Data Radio Bearer → the MAC scheduler enforces it.
QoS Flow vs DRB
QoS Flow is the core-network unit of differentiation; DRB is the radio bearer carrying it. Mapping is many-to-one — flows with similar requirements share a DRB — and the gNB decides. 3GPP specifies the mechanism, not the policy.
Protocol layers
What does each layer of the NR user plane do?
Going down from IP: SDAP maps QoS Flows to radio bearers and marks the QFI. PDCP handles ciphering, integrity protection, header compression, reordering and duplicate detection. RLC provides segmentation and, in acknowledged mode, ARQ. MAC does multiplexing, HARQ and scheduling. PHY does coding and modulation.
The Level 2 addition: SDAP is new in NR and exists because QoS Flows are new. In LTE the core network signalled a bearer per QoS class; in NR it signals flows, and something has to map them onto radio bearers. That something is SDAP.
Why does RLC have three modes?
Transparent mode does nothing — used for broadcast where there is no return path. Unacknowledged mode segments and reorders but does not retransmit, which suits latency-sensitive traffic that would rather lose a packet than wait. Acknowledged mode adds ARQ.
The follow-up worth pre-empting: if HARQ already retransmits, why does RLC need ARQ? Because HARQ feedback is a single bit and can itself be misread. A NACK misinterpreted as an ACK loses the block, and only RLC's sequence numbering catches it. HARQ handles the common case fast; RLC catches its residual errors.
What is PDCP duplication and when is it used?
The same PDCP PDU is sent over two legs — two carriers under carrier aggregation, or two nodes under dual connectivity — and the receiver discards the duplicate.
It doubles resource consumption for one packet, so it is a reliability tool, not a throughput one. The use case is URLLC, where the alternative to spending twice the resource is missing a deadline.
Mobility
Walk through a handover
The source gNB configures measurements; the UE reports when an event triggers. The source decides, and sends a Handover Request to the target over Xn. The target admits and returns a container. The source forwards it to the UE as an RRCReconfiguration with reconfigurationWithSync. The UE accesses the target with random access and confirms with RRCReconfigurationComplete. Path switch moves the user plane at the core.
The Level 2 addition: name the interruption. In classic handover the UE detaches from the source before attaching to the target, so there is a gap of a few tens of milliseconds. Mention DAPS — where the UE keeps the source link while establishing the target — as the mechanism that removes it.
What are the measurement events?
A1 serving becomes better than threshold. A2 serving becomes worse than threshold. A3 neighbour becomes better than serving by an offset. A4 neighbour becomes better than threshold. A5 serving worse than one threshold while neighbour better than another. B1 and B2 are the inter-RAT equivalents.
A3 drives most intra-frequency handovers. A2 commonly triggers the measurement of other frequencies — no point burning power measuring alternatives while the serving cell is fine.
What is conditional handover and what problem does it solve?
The UE is given the target configuration in advance along with an execution condition, and it performs the handover itself when the condition is met.
The problem it solves is the vulnerability window: in classic handover the command is sent over a radio link that is, by definition, degrading. If the command is lost, the UE fails and must re-establish. Conditional handover delivers the command while the link is still good and defers execution — which is why it matters most at high speed and at mmWave.
Troubleshooting scenarios
These test whether you reason systematically or guess. The winning approach is always the same shape: establish what "bad" means, then work down the stack eliminating layers.
Low throughput despite strong RSRP
The classic, because it forces you past signal strength.
RSRP measures received power, not quality. Start by separating the candidates:
Interference — good RSRP with poor SINR. Check SINR and CQI; RSRP high and SINR low points at a neighbour cell or external interference.
High BLER — check the retransmission rate. Persistent NACKs mean MCS is too aggressive for actual conditions.
Rank limitation — line-of-sight or a poorly conditioned channel collapses MIMO rank. Strong signal, one layer.
Carrier aggregation not active — SCells configured but not activated, or a band combination the UE doesn't support.
Cell load — you're getting a small share of a busy cell. Check PRB utilisation.
Not the radio at all — TCP behaviour, server capacity, transport congestion. Test with a local iperf server to rule the radio in or out.
High BLER
Order matters here. First check whether it's initial BLER (link adaptation too aggressive — often correct, since ~10% is a normal target) or residual BLER after retransmissions (a real problem).
Then: SINR and interference; CQI reporting delay against UE speed, since a fast UE's reports go stale; uplink power headroom if it's UL BLER; and timing advance or synchronisation issues.
High latency
Separate the contributors: scheduling delay (uplink grant round trip — check whether configured grants would help), HARQ retransmissions, DRX configuration (a long cycle adds latency by design), RRC state (resuming from IDLE costs 50–100 ms), UPF placement (central UPF for an edge application), and transport.
The strong move is asking which latency — user-plane one-way, round trip, or control-plane setup. They have different causes.
System design scenarios
Senior interviews lean here. The structure that works: requirements → constraints → architecture → trade-offs → how you'd validate.
Design a 5G network for a smart factory
Start by extracting requirements rather than proposing a solution. Traffic types (motion control, AGVs, cameras, sensors), latency and reliability targets per type, coverage area and RF environment, mobility, and who operates it.
Then the architecture: private 5G SA — you need the 5G Core for slicing and QoS. Local UPF on premises for latency and data sovereignty. Slices or at minimum differentiated QoS per traffic class: delay-critical GBR for motion control, GBR for AGVs, non-GBR for cameras. TDD pattern weighted for the uplink-heavy camera traffic, which is the detail most candidates miss. Redundancy via PDCP duplication or dual connectivity for the critical loops.
Then the trade-offs: URLLC capacity is expensive and comes out of eMBB capacity; RedCap for sensors reduces device cost but has worse link budget; private spectrum availability varies by country.
Design for massive IoT
Requirements: device count, message size and frequency, battery target, mobility, coverage including deep indoor.
Architecture: RedCap or NB-IoT depending on data rate — and being able to say which and why is the whole question. RedCap for anything needing megabits, NB-IoT for hundreds of bytes on a decade-long battery. Then eDRX and RRC_INACTIVE for power, an mMTC slice, and capacity planning driven by signalling load rather than throughput.
That last point is the insight: for massive IoT, the constraint is control-plane capacity, not user-plane.
Design a high-capacity stadium network
Extreme density, uplink-heavy (everyone uploading video), predictable peaks, severe interference from many small cells.
Architecture: dense small cells with massive MIMO for spatial reuse — this is exactly the load condition where MU-MIMO pays. Carrier aggregation across all available spectrum. TDD pattern adjusted for uplink. Careful cell-edge planning and inter-cell interference coordination. Consider mmWave for the seating bowl where line of sight exists.
Open RAN and 5G Advanced
What is Open RAN, and how does it differ from O-RAN?
Open RAN is the concept — disaggregation with open, interoperable interfaces. O-RAN Alliance is the organisation writing the specifications for it. vRAN is a separate axis: running RAN software on general-purpose hardware, which can be done without any open interfaces at all.
Getting these three apart is the answer that signals you have worked with it rather than read about it.
What does the RIC do, and why are there two?
The near-real-time RIC hosts xApps and acts on 10 ms to 1 s timescales over E2 — handover decisions, per-UE resource control. The non-real-time RIC hosts rApps inside the SMO, acting above 1 second — training models, setting policy, planning.
The split exists because the useful timescales differ by orders of magnitude. Anything that needs data aggregated across many cells over hours cannot also run inside a scheduling loop.
What is RedCap, and who is it for?
A reduced-capability NR device class from Release 17: narrower bandwidth, fewer antennas, fewer MIMO layers, optionally half-duplex. It fills the gap between NB-IoT and full eMBB.
The strong answer names the target segments — industrial sensors, wearables, video surveillance — and notes the coverage consequence: fewer receive antennas costs roughly 3 dB, so RedCap devices need denser deployment than smartphones for equivalent service.
What is Non-Terrestrial Networking?
3GPP support for satellite access, from Release 17. The hard problems are propagation delay — up to hundreds of milliseconds for GEO — and Doppler from fast-moving LEO satellites.
The specification adapts timers, extends the random access response window, and allows HARQ feedback to be disabled per process, because waiting for feedback across a GEO round trip would idle the link almost entirely.
Questions to ask them
Interviews run both ways, and the questions you ask place you as precisely as the ones you answer.
"What does your deployment look like — SA or NSA, and which bands?" Practical, and the answer tells you what you would actually work on.
"Where does your team sit relative to the standards work?" Distinguishes an organisation implementing specifications from one contributing to them.
"What is the hardest problem the team has open right now?" The best question in any technical interview. You learn what the work is really like, and asking it signals that you expect the work to be hard.
"How do you test?" For anything protocol-related, the test infrastructure tells you more about engineering maturity than the architecture does.
The compact cheat sheet
Numerology: Δf = 2^µ × 15 kHz. Wider spacing = shorter slot, shorter CP.
HARQ: soft combining, IR with RV0–RV3, up to 16 processes, K1/K2 signalled per transmission.
PF metric: instantaneous rate ÷ average throughput. Exploits multi-user diversity.
QFI identifies · 5QI characterises · ARP decides admission and preemption.
AMF = device · SMF = session · UPF = packets. N4 carries PFCP: PDR, FAR, QER, URR.
S-NSSAI = SST + optional SD. Requested ≠ allowed. Selection ≠ reservation.
CU = RRC/SDAP/PDCP · DU = RLC/MAC/PHY. Split sits where hard real-time begins.
MU-MIMO, not beamforming, is where massive MIMO capacity comes from.
RRC_INACTIVE: context retained at the anchor gNB; the core never knows.
SA vs NSA: slicing, RedCap, VoNR, URLLC and edge UPF all require the 5G Core.
Final strategy
Prepare ten questions properly rather than fifty superficially. Depth on numerology, HARQ, the QoS chain, the core function split, scheduling, RRC states, slicing, beamforming, CA versus DC, and one troubleshooting scenario will carry most interviews.
Practise the three-level structure out loud. It feels artificial until it doesn't, and then it's automatic.
Prepare one end-to-end walkthrough — a packet from application to air interface and back. Being able to narrate that cleanly demonstrates more than any individual definition.
Have a real example ready — something you debugged, designed, or got wrong. Interviewers remember the specifics far longer than the definitions.
And when you hit something you don't know, say so, then reason out loud about how you'd approach it. That's frequently the most informative answer you'll give.
Further reading
Deep-dives referenced throughout: 5G NR · gNB · 5G Core · AMF/SMF/UPF · PDU Sessions · QoS Flows · Network slicing · Scheduling · HARQ · RRC states · Beamforming · Massive MIMO · Carrier aggregation · Dual connectivity · RedCap · NSA vs SA
Related Articles
Non-Terrestrial Networks: 3GPP Satellites, Delay Budgets, and Direct-to-Device
How 3GPP adapted NR for satellites: LEO/MEO/GEO trade-offs, the timing advance and HARQ changes long delays forced, transparent versus regenerative payloads, and what direct-to-device really delivers.
What Is 5G NR? Numerology, Frequency Ranges, and the Flexible Frame
5G NR explained for engineers: scalable numerology, FR1 and FR2, the flexible slot structure, bandwidth parts, and what actually changed from LTE.
5G Scheduling Algorithms: Round Robin, Proportional Fair, and What Real Schedulers Do
How the gNB MAC scheduler decides who transmits: the classic algorithms, why Proportional Fair became the baseline, and how QoS-aware scheduling handles URLLC and eMBB together.