Integrated Sensing and Communication: Turning the Radio Network Into a Sensor

Every base station already transmits radio waves into an environment full of objects, and those objects already reflect some of that energy back. Today the network treats those reflections as multipath — something to equalise away.

Integrated Sensing and Communication is the proposal that we stop discarding that information and start using it. Same infrastructure, same spectrum, same waveform, second output: a picture of the physical environment alongside the data.

It's one of the more credible 6G differentiators, precisely because the physics has been understood since radar. The hard part isn't detecting reflections — it's doing so without giving up the throughput you were selling in the first place.

---

Why sensing belongs in the network

5G already delivers massive MIMO, beamforming, mmWave operation, high throughput, low latency, and advanced positioning. But the network's role stays fundamentally about moving information between endpoints.

6G research broadly proposes widening that role:

From a network that connects devices to one that perceives its environment.

The economic argument is infrastructure reuse. A 6G node already has antennas, RF chains, licensed spectrum, baseband processing, precise synchronisation, and backhaul. Dedicated sensing infrastructure — radar networks, camera systems — needs all of that built separately. If sensing can ride on what's already deployed, the marginal cost is mostly signal processing.

---

How radio reflections carry information

Communication and sensing look like different problems but share one resource: radio waves interacting with an environment.

Communication tries to recover a transmitted message despite the channel. Sensing tries to characterise the channel itself. Both depend on frequency, time, phase, amplitude, propagation, antennas, and signal processing — so the same receiver chain can, in principle, do both.

A reflected signal carries four kinds of information:

Time delay gives range. Doppler shift gives radial velocity. Angle of arrival gives direction. Amplitude and phase carry information about the object and the propagation path.

Combine them and you get a spatial description of what's out there.

Range from delay

Propagation delay maps directly to distance:

        c × Δt
R  =  ──────────
           2

The factor of two is because the signal makes a round trip. Straightforward — until you consider that resolving two objects a metre apart means resolving delays a few nanoseconds apart.

Velocity from Doppler

A moving object shifts the frequency of the reflection: approaching objects shift up, receding objects shift down.

This gives radial velocity — motion toward or away from the sensor. Tangential motion produces no Doppler shift at all, which is a real limitation for single-node sensing and one of the arguments for distributed sensing later on.

Angle from antenna arrays

A single antenna gives no directional information. An array does, because a wavefront arriving off-boresight reaches different elements at slightly different times.

More elements and wider aperture mean finer angular resolution — which is why massive MIMO matters so much here.

---

Why 6G makes this viable

Two developments turn a decades-old idea into something deployable.

Massive MIMO

Large arrays give both directional transmission and directional reception, and they let the system steer beams to scan an area rather than illuminating everything at once:

The same array that beamforms toward a UE can sweep a sensing beam across a sector. The hardware is already there.

Bandwidth

Range resolution — the ability to distinguish two objects at similar distances — depends on bandwidth:

              c
ΔR  ≈  ─────────────
           2 × B

Put numbers on it and the case becomes obvious. 100 MHz gives roughly 1.5 m resolution — enough to detect a vehicle, not to characterise it. 1 GHz gives around 15 cm — enough to distinguish a pedestrian from a cyclist. 10 GHz, plausible in sub-THz bands, gives about 1.5 cm.

That progression is why sensing is a 6G topic rather than a 5G one. The wide bandwidths available at mmWave and sub-THz frequencies are exactly what sensing resolution needs — and those bands' poor propagation matters less for sensing, which is inherently short-range anyway.

---

The resource problem

Here's the fundamental tension. Communication and sensing want the same time, frequency, power, and antenna resources.

The simplest approach is to partition them — dedicate some resources to sensing, the rest to communication:

Straightforward, but every resource given to sensing is throughput lost. This is what makes ISAC an economic question and not just a technical one.

The more ambitious approach is a shared waveform serving both purposes simultaneously:

This is where the research concentrates, and where the trade-offs get sharp. Communication waveforms are optimised for spectral efficiency and reliable demodulation. Sensing waveforms want good autocorrelation, unambiguous range-Doppler behaviour, and high SNR on weak reflections.

OFDM turns out to be a reasonable compromise — it's already deployed, and OFDM radar is well studied. But "reasonable compromise" means neither objective is optimal, and the scheduler ends up solving a multi-objective problem across throughput, latency, reliability, energy, sensing accuracy, range and angular resolution, coverage, and interference. That's a natural fit for AI-assisted radio resource management, and one reason ISAC and AI-native RAN research overlap heavily.

---

What ISAC is not

Not positioning

Positioning asks where is the UE? — and requires the target to be a cooperating device that transmits, receives, and reports measurements.

ISAC asks what is in the environment? The sensed object needs no radio at all.

Detecting a pedestrian, an unregistered drone, or a pallet in a warehouse aisle is categorically beyond what positioning can do, because none of those things are connected to the network.

Not quite radar

ISAC borrows radar's physics but inverts the design priorities. Radar is purpose-built for detection: waveform, hardware, and deployment all optimised for it. A cellular system is purpose-built for communication, with sensing added subject to not degrading that.

A dedicated radar will outperform ISAC on any single sensing metric. ISAC's advantage is ubiquity and cost — thousands of existing sites, already powered, already connected, already synchronised.

---

Distributed sensing changes the picture

Single-node sensing has hard limits: occlusion, tangential blindness, and weak returns from small or distant objects. Multiple nodes observing the same object fix most of that.

Each node sees a different aspect angle, which improves position accuracy through geometric diversity, resolves the tangential velocity problem, handles occlusion where one node's view is blocked, and improves detection reliability by combining weak returns.

This is where cellular infrastructure has an advantage that dedicated radar doesn't: the network is already a dense grid of synchronised, connected nodes. Multi-static radar normally requires purpose-built deployment. A cellular network is one by construction.

The catch is synchronisation. Bistatic and multistatic processing needs timing and phase coherence between nodes far tighter than communication requires. Getting there means better-than-PTP distribution, and it's one of the harder unsolved engineering problems in the field.

---

Monostatic sensing has a self-interference problem

The tidiest form of ISAC — a base station transmitting and listening for its own echo — runs into a physical obstacle that is easy to state and hard to solve.

The transmitted signal is enormously stronger than its own reflection. A target echo may arrive 100 to 150 dB below the transmission, and if the node is transmitting while listening, its own signal saturates the receiver front end. Everything after that point is measuring the transmitter, not the target.

Radar solves this with time separation: transmit a pulse, switch off, listen. That works because radar has no obligation to carry traffic. A base station transmitting in bursts to make listening windows is a base station not serving users during those windows.

The alternative is full-duplex operation with self-interference cancellation, in three stages: antenna isolation and placement, analogue cancellation before the low-noise amplifier, and digital cancellation after it. Together these can suppress well over 100 dB in laboratory conditions. Achieving it on a live base station with a moving environment, temperature drift and nonlinear amplifiers is another matter, and it is one of the main reasons early ISAC deployments favour bistatic geometries — one node transmits, a different node receives — where the problem largely disappears.

---

What a sensing measurement actually produces

It is worth being concrete about the output, because "the network senses the environment" conceals a great deal of signal processing.

The receiver correlates what it received against what was transmitted, producing a range-Doppler map: a two-dimensional surface with delay on one axis and frequency shift on the other. A reflecting object appears as a peak whose delay gives range and whose Doppler gives radial velocity.

Range resolution is set by bandwidth alone — roughly c/2B. At 100 MHz that is about 1.5 metres; at 1 GHz, 15 centimetres. This is why sensing performance improves so sharply with mmWave and sub-THz bandwidth.

Velocity resolution is set by observation time. Longer coherent integration separates closer velocities, but the target must stay coherent throughout, which limits how long is useful.

Angular resolution comes from the array. More elements, narrower beam, better bearing estimate.

Then comes the hard part. A raw map contains returns from everything: the target, walls, ground, vegetation, and the array's own sidelobes. Clutter suppression removes static returns, typically by subtracting a background estimate. Detection applies a threshold that adapts to local noise. Association and tracking link detections across time into object trajectories.

The output the network can act on is several processing stages removed from the radio measurement, and each stage is where accuracy is won or lost.

Weak reflections. Small objects at distance return very little energy, and the receiver must find it under noise, interference, hardware impairments, and self-interference from the node's own transmission — a full-duplex problem in itself.

Multipath. Urban and industrial environments are full of reflective surfaces, so a return may arrive by several paths.

Distinguishing a genuine target from a wall reflection is difficult, and it's precisely the information that communication systems have spent thirty years learning to suppress.

Data volume. A network sensing continuously generates enormous measurement volumes, driving requirements for edge processing, intelligent filtering, compression, and distributed inference.

Privacy. This is the one most likely to shape whether ISAC deploys at all. A network that detects people, movement, and activity — including through walls at some frequencies, and including people who never consented because they aren't subscribers — raises questions that no existing telecom privacy framework answers.

Communication networks have a clean model: you're a subscriber, you agreed to terms, your data is protected. Sensing breaks it entirely. The person walking past your building isn't a subscriber and has no relationship with the operator. Regulation here is likely to be as decisive as the technology.

---

Privacy is a design constraint, not a footnote

A network that can detect and track objects can detect and track people, and it does so without their participation, without a device, and without any moment at which consent could reasonably be obtained.

This is qualitatively different from the privacy questions mobile networks already face. Location of a subscriber is derived from a device that person chose to carry and a subscription they entered into. Sensing has neither property — a person walking past a base station is measured whether or not they have any relationship with the operator.

The technical mitigations are real but partial. Resolution limits can be set so the system distinguishes "a person is present" from "this specific person is present" — the coarser the measurement, the weaker the identification. On-node processing that emits only derived events rather than raw range-Doppler data reduces what can be reconstructed later. Retention limits shorten the window in which movement histories exist at all.

But gait is close to biometric, and a sufficiently high-resolution sensing system distinguishes individuals whether or not it was designed to. The regulatory position in most jurisdictions is unsettled, and it is likely to determine which ISAC use cases are commercially viable more decisively than any technical constraint.

The use cases that avoid the problem entirely — drone detection, industrial safety inside a controlled facility, infrastructure monitoring — are the ones likely to deploy first, and that ordering is driven by law rather than by engineering difficulty.

ISAC doesn't arrive from nowhere with 6G. 5G-Advanced already extends positioning accuracy, sidelink positioning, and RF sensing study work — Release 18 and beyond building the foundations.

The plausible near-term applications are the constrained ones: industrial environments, where the deployment is private, the space is controlled, the objects are known, and privacy is governed by employment rather than public policy. Drone detection around airports and critical infrastructure, where the target genuinely is uncooperative and the alternative infrastructure is expensive. Traffic and infrastructure monitoring, where aggregate flow data is useful and individual identification isn't needed.

The more ambitious visions — vehicle safety augmentation, ubiquitous environmental awareness feeding digital twins — depend on resolution, reliability, and regulatory clarity that don't exist yet.

The architectural shift is real regardless: the base station stops being purely a connectivity endpoint and becomes an environmental sensor with backhaul. Whether that becomes a mainstream network capability or a specialised vertical feature is the open question — and it'll be settled by economics and regulation at least as much as by physics.

---

Further reading

• 3GPP TR 22.837 — Study on integrated sensing and communication
• 3GPP TS 22.137 — Service requirements for integrated sensing and communication
• Hexa-X and Hexa-X-II — European 6G flagship projects, sensing work packages
• ITU-R M.2160 — IMT-2030 framework, covering integrated sensing and communication