CFAR Detection on Non-Coherently Integrated Angular Maps for mmWave Radio SLAM
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: We're looking at the paper: "CFAR Detection on Non-Coherently Integrated Angular Maps for mmWave Radio SLAM".
Dev: Here's the gist of the paper. Reliable multipath detection is essential for millimeter-wave radio simultaneous localization and mapping (SLAM), where propagation paths reflecting off the environment provide geometric constraints beyond the dominant line-of-sight component.
Rosa: The paper develops constant false alarm rate (CFAR)-based detectors for beam-swept bistatic integrated sensing and communications (ISAC).
Dev: System Model Details The system considers a bistatic mmWave link with a fixed BS and a mobile UE, focusing only on the 2D azimuth plane.
Rosa: The received signal at the k-th subcarrier and m-th OFDM symbol is given by equation (1).
Title and authors: Rosa: So we're looking at this paper called "CFAR Detection on Non-Coherently Integrated Angular Maps for mmWave Radio SLAM." It tackles how to reliably find those reflections off walls or objects in millimeter wave radio systems, which is super important for figuring out exactly where a device is using SLAM.
Dev: Yeah, it focuses on making the detection robust when you're dealing with those complex multipath environments, not just the direct line of sight signal. The authors are developing constant false alarm rate CFAR detectors specifically for beam-swept bistatic integrated sensing and communications, which means they are looking at both the transmitter and receiver simultaneously.
Taro: What does that mean in practice? Does this help us when the device is indoors where you can't see a clear path to a satellite or cell tower? I mean, real world conditions are messy.
Rosa: Exactly. The paper shows they start from conventional CFAR on the I/Q datacube but then modify it because the noise statistics after integration aren't what standard cell-averaging CFAR expects them to be. They find that the conventional threshold doesn't work well anymore because those power map cells end up looking like Gaussians.
Dev: That leads right into their main proposal, which is a modified detector for non-coherently integrated angular power maps. Instead of treating every single measurement separately, they integrate the data first and then set a threshold based on a different noise model.
Taro: So what's the practical difference when you move from that standard approach to this modified one? Is it just a small number improvement or something bigger?
Rosa: The results show some pretty tangible gains. Compared to using Singular Value Decomposition, their proposed CFAR approaches can reduce the overall positioning root mean squared error from zero point three zero meters down to between zero point two four and zero point two six meters <ref:2611.14250#pg2>.
Dev: That reduction is most noticeable in non-line-of-sight snapshots, which is where we expect the biggest challenges. The multi-cell noncoherent detector specifically lowers that positioning root mean squared error from zero point three zero meters to zero point two five meters overall, which is a solid improvement <ref:2611.14250#pg2>.
Taro: If you're dealing with those tricky indoor situations where direct line of sight is blocked, how does this help the system handle things when the environment misbehaves? What's the safety case here?
Rosa: The paper introduces a multi-cell cell under test formulation and a plus-shaped training window. These are techniques they use to reduce computational cost while still accounting for that directional structure of sidelobe energy along those angle axes.
Title and authors: Dev: And the computational efficiency is pretty impressive because non-coherent integration requires ten to the power of three less statistic evaluations compared to detecting every single received element individually, which means a much lower threshold time and total detector time for map formation.
Taro: That kind of reduction in processing load is huge when you are trying to run this in real time on a mobile device or a small embedded system. But what about the angle and delay estimation part? How precise are they getting those angles?
Rosa: For angle estimation, they fit a two dimensional Gaussian with a constant pedestal over a fixed square patch of cells centered on the peak in the power map to get the angle. They get their coarse delay from looking at the time domain I Q of that detected beam pair and then use a sub sample grid search for that finer delay detail.
Dev: That fine delay estimation is crucial because you need to filter out those weaker returns that are close in both angle and delay to a stronger peak, which is what the sidelobe filtering helps accomplish. They also mention they can tune the per-RE CFAR detector to get the best performance in a narrow band around ten to ten to the minus twelve, and their single CUT detector works well between ten to the minus eleven and ten to the minus thirteen <ref:2611.14250#pg2>.
Taro: So if I'm trying to set up my system, what's the practical implication of that tuning range? Does it mean I can dynamically adjust my detection threshold based on how noisy the signal is right then?
Rosa: Yes, that’s what they are showing. It gives you a method for dynamically selecting the optimal detection threshold based on noise conditions, which maximizes accuracy under varying multipath scenarios. The experimental results show their proposed detectors achieve an overall root mean squared error of zero point two four to zero point two six meters, better than the zero point three zero meter result from Singular Value Decomposition <ref:2611.14250#pg2>.
Dev: That performance holds across both line-of-sight and non-line-of-sight snapshots, which is what they show demonstrates their suitability for practical radio SLAM applications. The operating band between ten to the minus six and ten to the minus nine represents the practical working range for this noncoherent detector <ref:2611.14250#pg2>.
Taro: So to wrap up on what this paper means, it’s about making sure that when we are using millimeter wave systems for simultaneous localization and mapping in real environments, we can trust our positioning even when there's a lot of clutter and reflections.
Title and authors: Rosa: Right. The core idea of this paper, "CFAR Detection on Non-Coherently Integrated Angular Maps for mmWave Radio SLAM," is that standard CFAR assumptions break down when you integrate the data noncoherently, so they developed a modified detector that handles those different noise statistics better than before.
Dev: And they did it efficiently by using a multi-cell formulation and a plus-shaped window to keep the computational overhead low while still capturing the necessary sidelobe information. The key takeaway is that this method provides significantly better positioning accuracy in non-line-of-sight scenarios compared to singular value decomposition based detectors.
Taro: For me, what’s interesting is how they handle those estimation steps—the angle and delay fitting—by using a two dimensional Gaussian with a constant pedestal, which lets them refine the angles even if the true angle isn't perfectly on a grid point.
Rosa: Right. And for timing, they use a sub sample grid search to get that fine delay estimate by maximizing coherent correlation of the least squares PRS channel estimate on those detected beam pairs after compensating for the coarse delay. It’s all tied together to give you better localization accuracy in these complex radio maps.
Dev: So, while this paper shows strong performance, it's important to remember their limitation that they focused only on the 2D azimuth plane for this work and evaluated it on experimental sixty GHz indoor measurements using a snapshot-SLAM.
Taro: And that means if you try to apply this to a different environment or frequency band, you might need to re-evaluate those specific noise statistics because the paper suggests they are unique to that setup.
Rosa: Exactly. The paper gives us a solid tool for improving detection in mmWave radio SLAM, but we need to keep an eye on how these noncoherent integration methods scale up when you move from small indoor snapshots to longer-term mapping problems.
Dev: So, the big picture here is that this paper provides a way to achieve higher accuracy in localization by using smarter statistical detection techniques tailored for the specific noise properties of integrated sensing and communications systems.
Taro: It sets a path forward for making these radio SLAM solutions more robust against real-world environmental challenges, especially when dealing with those challenging non-line-of-sight conditions.
Rosa: Yeah, so that’s our take on the paper "CFAR Detection on Non-Coherently Integrated Angular Maps for mmWave Radio SLAM." It’s a solid piece of work for making those radio systems more reliable.
The paper's summary: Rosa: So, to recap, this paper is about taking that standard CFAR detection method we use and making it work for these mmWave systems where the signal paths are all mixed up in a messy way because of reflections off the environment.
Dev: Exactly. It’s focusing on how you detect those reflection signals when you’re not keeping everything perfectly coherent, which is a big problem when you're trying to map out an area using radio SLAM.
Rosa: The authors found that the standard way of setting your threshold just doesn't work anymore because the noise statistics change once all those different paths are averaged together. It assumes a certain type of noise that isn't actually what you get in this integrated sensing setup.
Dev: And they propose a new method, a modified detector for these non-coherently integrated maps, which handles those different noise stats correctly by adjusting the threshold differently. They also added this multi-cell extension to account for the fact that one reflection might show up across several adjacent beams at once.
Rosa: What’s really interesting is their performance numbers. They compared it against a singular value decomposition detector, and their modified method cut the positioning error down from about zero point three zero meters to around zero point two four or zero point two six meters.
Dev: That reduction is most significant when the system can't see a direct path, those non-line-of-sight snapshots. It shows that this new statistical approach gives the AI a much better sense of where it is, even when the environment is cluttered and confusing.
Rosa: And they also made it computationally cheaper, which for real systems running in real time matters a lot. They introduced this multi-cell formulation and a plus-shaped training window to keep the math simple while still getting that accuracy boost.
Dev: That computational efficiency is pretty impressive because they found that their non-coherent integration requires about one thousand times fewer statistic evaluations than checking every single element separately, which means less time spent waiting for the detection to finish.
Rosa: But then you have the angle estimation part, which is where you figure out where the signal came from. They use a two dimensional Gaussian fit over a patch of cells to get the angle, and they get their fine delay by doing this sub sample grid search on those detected beam pairs.
Dev: That sub sample grid search is how they filter out those weaker returns that are just sitting close in both angle and time, which helps clean up the final localization estimate. They also showed you can tune the detector threshold for a really specific noise band, around ten to ten to the minus twelve hertz, depending on whether you're using a standard method or their single cut version.
Rosa: So basically, it’s a more robust way to find those reflections so that the SLAM system doesn't get lost in the noise when things are complex and we can use it even when direct line of sight is gone.
Dev: It’s a solid tool for practical radio SLAM, but remember they only focused on the two dimensional azimuth plane for this work and tested it using experimental measurements at sixty gigahertz. So, applying this to a different frequency or a different kind of environment might require re-evaluating those specific noise conditions.
Rosa: Right. It’s good to have a tool that works well in these controlled indoor setups, but we still need to see how it holds up when you take it out into the real world where the physics gets even more unpredictable and complex.
The paper's improvements: Rosa: So, we’re looking at how they suggest using their new detection method to actually get better results in practice, beyond just the numbers on paper.
Dev: Right. They aren't just saying it’s theoretically sound; they’re showing how this actually changes the system when you run trials outside the lab conditions we usually test in.
Rosa: One big thing they suggest is that you can tune your detection threshold dynamically, depending on how noisy the signal looks at that specific moment. They found a sweet spot around ten to ten to the minus twelve hertz for one of their methods, but they showed you can adjust it based on your environment.
Dev: And they also have this idea about using a single cut detector between ten to the minus eleven and ten to the minus thirteen hertz, which lets you pick the best method depending on what kind of noise you’re dealing with in real time.
Rosa: That means instead of just picking one fixed number for your threshold, you can let the system adapt to whether it’s a quiet day or a really noisy snapshot. It makes the AI system more resilient when things get messy.
Dev: And they also talked about how this helps with angle estimation and delay refinement, which is key for localization accuracy even when the true angle isn't perfectly aligned with one of the grid points in their map.
Rosa: So, if you’re using an AI system that needs to pinpoint a location precisely, this suggests you can get better angular estimates even when your initial power map cells aren't perfect squares.
Dev: It also gives them a way to filter out those weak reflections that are close in both angle and time, which keeps the final positioning estimate cleaner and more focused on the strongest returns.
Rosa: This whole approach is really about making the radio SLAM system more robust when it encounters those hard, non-line-of-sight situations we see all the time in real building environments.
Dev: It means that for a controls engineer, it gives you a method where you can build in some flexibility to handle unexpected signal behavior without having to completely overhaul your entire detection pipeline every time.
Rosa: So, the implication is that we move toward radio SLAM systems that aren't just good in perfect test rooms but are actually dependable when deployed on mobile robots or drones navigating real-world indoor spaces.
Conclusion: Rosa: So, to wrap up, this paper is about how they modified CFAR detection for non-coherently integrated angular power maps in mmWave radio SLAM systems and showed that this approach improves positioning accuracy compared to other methods.
Dev: Yeah, it’s a solid piece of work showing a way to get better localization when the signal paths are messy and you can't rely on perfect line of sight.
Rosa: The main implication is that we can build radio SLAM systems that are more reliable in real-world indoor settings, not just in controlled test labs.
Dev: It means for real-time systems, we have a better safety case because the error reduction is substantial when dealing with those non-line-of-sight snapshots.
Taro: I think it’s interesting how they show this works across both line of sight and non-line-of-sight scenarios, which really shows its versatility for autonomy applications.
Rosa: Exactly, and the computational efficiency gain means we can actually run this kind of detection on mobile platforms without crippling the processor.
Dev: But there are limitations, so remember they only focused on the two dimensional azimuth plane for this specific work and tested it at sixty gigahertz measurements. So, if you want to use it elsewhere, you’ll need to check those noise statistics again.
Taro: That makes sense; we have to be careful about applying results across different frequency bands because the noise characteristics change depending on what you're sensing.
Rosa: True, but this paper definitely gives us a much more robust statistical tool for dealing with complex multipath environments in mmWave systems.
Dev: It’s a step forward for making these radio SLAM solutions more practical and accurate under challenging conditions.
Paolo Tosi, Ossi Kaltiokallio, Marcus Henninger, Silvio Mandelli, Mikko Valkama
Nokia Bell Labs · Karlsruhe Institute of Technology (KIT) · Tampere University
eess.SP
Submitted: 2026-11-10
Updated: 2026-11-10
Comments: 6 pages, 4 figures, 1 table. This work has been submitted to the IEEE for possible publication
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
Importance score: 83/100
The gist: Reliable multipath detection is essential for millimeter-wave radio simultaneous localization and mapping (SLAM)…
Key concepts
- CFAR Detection
- CFAR is a technique used to set detection thresholds for signal strength maps to maintain a constant false alarm rate, meaning the probability of incorrectly detecting noise remains steady. The paper modifies this standard method because the noise statistics in their integrated power maps are different from what standard CFAR assumes.
- Non-Coherent Integration
- This refers to combining multiple signal measurements (like those from different subcarriers or symbols) without fully coherently processing them first. The proposed detector is specifically designed to handle the unique noise characteristics that arise when integrating these non-coherent measurements, which standard detectors fail to model correctly.
- Multi-Cell Block / Plus-Shaped Window
- These are modifications made to the CFAR detection region. The multi-cell block accounts for path returns occupying multiple neighboring beam pairs, while the plus-shaped window addresses the directional structure of sidelobe energy along the angle axes, leading to better performance.
- Angle and Delay Estimation
- This process determines where a detected signal is coming from (angle) and when it arrived (delay). The system fits a 2D Gaussian shape to the power map to find the angle, and then uses sub-sample grid searching on time-domain data to precisely estimate the delay.
Terminology
Summary
Reliable multipath detection is essential for millimeter-wave radio simultaneous localization and mapping (SLAM), where propagation paths reflecting off the environment provide geometric constraints beyond the dominant line-of-sight component. p. 0 CFAR Detection on Non-Coherently Integrated
Angular Maps for mmWave Radio SLAM
Paolo Tosi∗†, Ossi Kaltiokallio‡ Marcus Henninger∗
, Silvio Mandelli∗
, and Mikko Valkama‡
∗Nokia Bell Labs, Stuttgart, Germany, †Karlsruhe Institute of Technology (KIT)
‡Tampere University, Tampere, Finland
Abstract—Reliable multipath detection is essential for
millimeter-wave radio simultaneous localization and mapping
(SLAM), where propagation paths reflecting off the environment
provide geometric constraints beyond the dominant line-ofsight component. This paper develops constant false alarm
rate (CFAR)-based detectors for beam-swept bistatic integrated
sensing and communications (ISAC). CFAR Detection on Non-Coherently Integrated
Angular Maps for mmWave Radio SLAM
Paolo Tosi∗†, Ossi Kaltiokallio‡ Marcus Henninger∗
, Silvio Mandelli∗
, and Mikko Valkama‡
∗Nokia Bell Labs, Stuttgart, Germany, †Karlsruhe Institute of Technology (KIT)
‡Tampere University, Tampere, Finland
Abstract—Reliable multipath detection is essential for
millimeter-wave radio simultaneous localization and mapping
(SLAM), where propagation paths reflecting off the environment
provide geometric constraints beyond the dominant line-ofsight component. This paper develops constant false alarm
rate (CFAR)-based detectors for beam-swept bistatic integrated
sensing and communications (ISAC).Where the paper says it
Where the paper says it
System Model Details
The system considers a bistatic mmWave link with a fixed BS and a mobile UE, focusing only on the 2D azimuth plane. The received signal at the k-th subcarrier and m-th OFDM symbol is given by equation (1). p. 1 The signal propagates in a multipath
environment with 𝑁 propagation paths. For the 𝑖
th TX beam
and 𝑗
th RX beam, the received signal at the 𝑘
th subcarrier and
𝑚
th OFDM symbol writes as
𝑦
𝑖, 𝑗
𝑘,𝑚 = w
𝐻
RX, 𝑗
H𝑘,𝑚w
∗
TX,𝑖𝑥
𝑖, 𝑗
𝑘,𝑚 + n
𝑖, 𝑗
𝑘,𝑚
, (1)
where wTX,𝑖 ∈ C
𝑁TX and wRX, 𝑗 ∈ C
𝑁RX are the TX and
RX beamformers, respectively, 𝑥𝑘,𝑚 ∈ C is the transmitted
PRS sample with |𝑥𝑘,𝑚| = 1∀𝑘, 𝑚, and n
𝑖, 𝑗
𝑘,𝑚 ∈ C
𝑁RX is
the additive white Gaussian noise (AWGN) at each RX
antenna element. For the 𝑖
th TX beam
and 𝑗
th RX beam, the received signal at the 𝑘
th subcarrier and
𝑚
th OFDM symbol writes as
𝑦
𝑖, 𝑗
𝑘,𝑚 = w
𝐻
RX, 𝑗
H𝑘,𝑚w
∗
TX,𝑖𝑥
𝑖, 𝑗
𝑘,𝑚 + n
𝑖, 𝑗
𝑘,𝑚
, (1)
where wTX,𝑖 ∈ C
𝑁TX and wRX, 𝑗 ∈ C
𝑁RX are the TX and
RX beamformers, respectively, 𝑥𝑘,𝑚 ∈ C is the transmitted
PRS sample with |𝑥𝑘,𝑚| = 1∀𝑘, 𝑚, and n
𝑖, 𝑗
𝑘,𝑚 ∈ C
𝑁RX is
the additive white Gaussian noise (AWGN) at each RX
antenna element. The transmitted signal propagates through
the multipath environment, whose effect is encoded in the
effective spatial multipath channel matrix H𝑘,𝑚 ∈ C
𝑁TX×𝑁RX
defined as [11, eq. In the former, both components are approximated by
independent N (0, 𝜎2
/2) densities, consistent with the AWGN
model. 2) Received BRSRP power
For each TX-RX beam pair (𝑖, 𝑗), the non-coherent power
map 𝛽 ∈ R
𝑁TX×𝑁RX×1
is formed by averaging over the REs as
𝛽𝑖, 𝑗 =
1
𝑁RS
∑︁
(𝑘,𝑚) ∈MRS
𝑣
𝑖, 𝑗
𝑘,𝑚, (6)
where 𝑣
𝑖, 𝑗
𝑘,𝑚 is the square law sample of (3) at AoD index 𝑖,
AoA index 𝑗, subcarrier 𝑘 and symbol 𝑚. 1) Raw IQ samples
The square law detector sample writes as
𝑣
𝑖, 𝑗
𝑘,𝑚 =
𝑦
𝑖, 𝑗
𝑘,𝑚
2
= ℜ𝔢{𝑦
𝑖, 𝑗
𝑘,𝑚}
2 + ℑ𝔪{𝑦
𝑖, 𝑗
𝑘,𝑚}
2
. (3)
When only AWGN is present, the in-phase and quadrature
components of 𝑦
𝑖, 𝑗
𝑘,𝑚 are independent zero-mean Gaussian
random variables of equal variance
ℜ𝔢{𝑦
𝑖, 𝑗
𝑘,𝑚} ∼ N (0, 𝜎2
/2), ℑ𝔪{𝑦
𝑖, 𝑗
𝑘,𝑚} ∼ N (0, 𝜎2
/2), (4)
so that a single square-law sample is exponentially distributed
with PDF
𝑝(𝑣) =
1
𝜎2
𝑒
−𝑣/𝜎2
(5)
and mean E[𝑣] = 𝜎
2
, i.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
CFAR Detection Methods
The paper introduces a modified detector for non-coherently integrated angular power maps where the post-integration noise statistics differ from those assumed by standard cell-averaging CFAR. The conventional threshold assumes exponentially distributed single-look samples, but the resulting power-map cells are approximately Gaussian and the conventional threshold no longer holds. p. 0 Its conventional threshold assumes exponentially distributed single-look
samples, so it can be applied to each positioning reference
signal (PRS) resource element (RE) before integration, at
the cost of one detection test per RE. Non-coherent integration avoids this cost, but the resulting power-map cells are
approximately Gaussian and the conventional threshold no
longer holds. To overcome this limitation, this paper develops
a CFAR test for multipath detection in beam-swept angular
maps that accounts for the post-integration Gaussian statistics. Because a path return occupies multiple neighbouring beam
pairs, the test region is extended to a multi-cell block, so
the statistic is evaluated once per block rather than once per
bin. A detection is stored at the strongest cell of the cell
under test (CUT) block, retaining the angular resolution of the
single-cell test. In addition, a plus-shaped CFAR window is
proposed, to account for the directionally structured sidelobe
energy along the AoA/AoD axes.Where the paper says it
Where the paper says it
Where the paper says it
Performance Gains
Compared with a singular value decomposition (SVD)-based detector, the proposed CFAR approaches reduce the overall positioning RMSE from 0.30 m to 0.24-0.26 m. The largest improvement is observed in non-line-of-sight snapshots. p. 0 Compared with a singular
value decomposition (SVD)-based detector, the proposed CFAR
approaches reduce the overall positioning RMSE from 0. 26 m, with the largest improvement in non-line-of-sight
snapshots. The proposed method
is evaluated on experimental PRS measurements from the
Campus Arena at Tampere University and compared with
the SVD-based BRSRP estimator of [8], using the same
downstream snapshot-SLAM backend of [9]. Multi-CUT noncoherent CFAR lowers positioning root mean squared error
(RMSE) from 0. The proposed methods
are evaluated on experimental 60 GHz indoor measurements
using a snapshot-SLAM estimator. Compared with a singular
value decomposition (SVD)-based detector, the proposed CFAR
approaches reduce the overall positioning RMSE from 0.Where the paper says it
Where the paper says it
Where the paper says it
Computational Efficiency
The computational cost is reduced by introducing a multi-cell cell-under-test formulation and a plus-shaped training window. Non-coherent integration lowers the number of statistic evaluations compared to per-RE detection, leading to a considerably lower threshold time and total detector time including map formation. p. 5 75×103
, 𝑇thr = 19. In terms of
computational complexity, non-coherent integration requires
103
less statistic evaluations than per-RE detection, leading to
a considerably lower threshold time and total detector time
including map formation. 75×103
, 𝑇thr = 19. In terms of
computational complexity, non-coherent integration requires
103
less statistic evaluations than per-RE detection, leading to
a considerably lower threshold time and total detector time
including map formation.Where the paper says it
Where the paper says it
Angle and Delay Estimation
Angle estimation involves fitting a 2D Gaussian with a constant pedestal to the power map over a fixed square patch centered on the retained peak. The coarse delay is obtained from the time-domain I/Q of the detected beam pair, and the fine delay is obtained by a sub-sample grid search. p. 4 The
fine delay 𝜏𝑓 ,𝑛 appears as the linear phase 𝑒
− 𝑗2𝜋𝑘Δ 𝑓 𝜏 across
subcarriers. It is obtained by a sub-sample grid search that
maximises the coherent correlation of the least-squares PRS
channel estimate on the detected beam pair after compensating
for the coarse delay. It is obtained by a sub-sample grid search that
maximises the coherent correlation of the least-squares PRS
channel estimate on the detected beam pair after compensating
for the coarse delay. Sidelobe filtering then discards weaker
returns that are close in angle and delay to a stronger peak.Where the paper says it
Where the paper says it
Experimental Results
The proposed detectors achieved an overall RMSE of 0.24–0.26 m, improving over the 0. p. 4 The CFAR detectors
achieve an overall RMSE of 0. 26 m, improving over the
0. The beam reference signal received
power (BRSRP)-based method in [8] addresses this limitation
by operating directly on the measured AoA/AoD power
map. The rank-1 singular value decomposition (SVD)-based
approach provides robustness against sidelobes and noise. A square training region
does not consider the angular return structure and sideobes
may mask weak returns. The plus-shaped window preserves
the contrast with the noise reference by selecting the larger
of the two per-strip statistics in (16), i. Below 10−9
, NLOS
error rises again, as weak multipath returns carrying geometric
information fall below threshold, and the SLAM estimator is
unable to solve the multipath problem. Between 10−6
and 10−9
both of its curves are nearly flat, so this represents the practical
working band.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Conclusion
In this paper we formulated conventional CFAR detection on the I/Q datacube for mmWave bistatic ISAC and derived a modified detector for non-coherently integrated angular power maps, including a multi-CUT extension for reduced complexity. The proposed detector is also able to detect weak multipath returns that are fundamental for solving the SLAM problem accurately. p. 5 On experimental data, the proposed detectors
reduced the overall positioning RMSE from the SVD baseline
by approximately 20%, with the largest improvement in NLOS
conditions. The proposed detector is also able to detect weak
multipath returns that are fundamental for solving the SLAM
problem accurately.Where the paper says it
Acknowledgments
Paolo Tosi has been supported by the European Commission through the ISLANDS project (grant agreement no. 101120544). p. 5 ACK NOW LEDG MEN TS
Paolo Tosi has been supported by the European Commission through the ISLANDS project (grant agreement no. 101120544).Where the paper says it
References
[1] J. Wang et al. p. 5 REFERENC ES
[1] J. Wang et al. Wang et al. , “Integrated sensing and communication: Enabling
techniques, applications, tools and data sets, standardization, and future
directions,” IEEE Internet Things J.Where the paper says it
Where the paper says it
[5] J. Salmi, A. Richter, and V. p. 5 93–107,
2018. Richter, and V. Richter, and V. Koivunen, “Detection and tracking of mimo
propagation path parameters using state-space approach,” IEEE Trans. Koivunen, “Detection and tracking of mimo
propagation path parameters using state-space approach,” IEEE Trans. Signal Process. Signal Process. 1538–1550, Apr.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[8] E. Rastorgueva-Foi et al. p. 5 Honolulu, HI,
USA: IEEE, Sep. Rastorgueva-Foi et al. Rastorgueva-Foi et al. , “Millimeter-wave radio SLAM: End-to-end
processing methods and experimental validation,” IEEE J. , “Millimeter-wave radio SLAM: End-to-end
processing methods and experimental validation,” IEEE J. Areas
Commun. Areas
Commun. 2550–2567, 2024.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[9] O. Kaltiokallio et al. p. 0 CFAR Detection on Non-Coherently Integrated
Angular Maps for mmWave Radio SLAM
Paolo Tosi∗†, Ossi Kaltiokallio‡ Marcus Henninger∗
, Silvio Mandelli∗
, and Mikko Valkama‡
∗Nokia Bell Labs, Stuttgart, Germany, †Karlsruhe Institute of Technology (KIT)
‡Tampere University, Tampere, Finland
Abstract—Reliable multipath detection is essential for
millimeter-wave radio simultaneous localization and mapping
(SLAM), where propagation paths reflecting off the environment
provide geometric constraints beyond the dominant line-ofsight component. This paper develops constant false alarm
rate (CFAR)-based detectors for beam-swept bistatic integrated
sensing and communications (ISAC). Kaltiokallio et al. , “Robust Snapshot Radio SLAM,” IEEE Transactions on Vehicular Technology, vol. , “Robust Snapshot Radio SLAM,” IEEE Transactions on Vehicular Technology, vol. 8460–8465, May
2025.Where the paper says it
Where the paper says it
Where the paper says it
[10] “NR; physical channels and modulation,” 3GPP, Technical Specification (TS) 38. p. 5 8460–8465, May
2025. [10] “NR; physical channels and modulation,” 3GPP, Technical Specification
(TS) 38. [10] “NR; physical channels and modulation,” 3GPP, Technical Specification
(TS) 38. 211, 2025, version 19.Where the paper says it
Where the paper says it
[7] C. Qin, J. A. Zhang, X. Huang, and Y. p. 5 831–846, Oct. Huang, and Y. Huang, and Y. Guo, “Angle-of-arrival
acquisition and tracking via virtual subarrays in an analog array,”
in Proc. Guo, “Angle-of-arrival
acquisition and tracking via virtual subarrays in an analog array,”
in Proc. IEEE 90th Veh. (VTC-Fall). Honolulu, HI,
USA: IEEE, Sep.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[11] R. W. Heath, N. Gonzalez-Prelcic, S. p. 5 211, 2025, version 19. Gonzalez-Prelcic, S. Gonzalez-Prelcic, S. Roh, and A. Roh, and A. ´
Sayeed, “An overview of signal processing techniques for millimeter
wave mimo systems,” IEEE J. Heath, “Channel
estimation and hybrid precoding for millimeter wave cellular systems,”
IEEE J. Topics Signal Process. Topics Signal Process. 436–453, 2016.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[12] M. A. Richards, Fundamentals of radar signal processing, second edition ed. p. 5 436–453, 2016. Richards, Fundamentals of radar signal processing, second
edition ed. Richards, Fundamentals of radar signal processing, second
edition ed. McGraw-Hill Education, 2014.Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Non-coherent integration detector can reduce overall positioning RMSE from 0.30 m to 0.24-0.26 m, with the largest improvement in non-line-of-sight snapshots compared to a singular value decomposition (SVD)-based detector. This allows the AI system to achieve significantly more accurate localization in complex indoor environments where direct line-of-sight is unavailable.
-
The proposed multi-cell cell under test formulation and plus-shaped training window reduce computational cost while accounting for sidelobe structure, demonstrating suitability for practical multipath radio SLAM. This enables the system to maintain high positioning accuracy even when operating with reduced computational resources on beam-swept angular maps.
-
The per-RE CFAR detector can be tuned to obtain the best performance in a narrow band around 10 to 10-12, and the single-CUT detector between 10-11 and 10-13. This provides a method for dynamically selecting the optimal detection threshold based on noise conditions, maximizing accuracy under varying multipath scenarios.
-
The system can perform angle estimation by fitting a two dimensional Gaussian with a constant pedestal over a fixed square patch of cells to the power map, allowing for the refinement of angles using nonlinear least squares. This capability enables precise localization even when the true angle does not align exactly with a grid point, as the estimation error is at most half a cell.
-
The AI system can estimate time of arrival by performing a sub sample grid search that maximizes coherent correlation of the least squares PRS channel estimate on detected beam pairs after compensating for coarse delay. This allows for fine delay estimation and sidelobe filtering, discarding weaker returns that are close in angle and delay to stronger peaks.
Abstract
Reliable multipath detection is essential for millimeter-wave simultaneous localization and mapping (SLAM), where propagation paths reflecting off the environment provide geometric constraints beyond the dominant line-of-sight component. This paper develops constant false alarm rate (CFAR)-based detectors for beam-swept bistatic integrated sensing and communications (ISAC). Starting from conventional CFAR on the I/Q datacube, we derive a modified detector for non-coherently integrated angular power maps, where the post-integration noise statistics differ from those assumed by standard cell-averaging CFAR. To improve practicality, we introduce a multi-cell cell-under-test formulation and a plus-shaped training window that reduces computational cost while accounting for sidelobe structure. The proposed methods are evaluated on experimental 60 GHz indoor measurements using a snapshot-SLAM estimator. Compared with a singular value decomposition (SVD)-based detector, the proposed CFAR approaches reduce the overall positioning RMSE from 0.30 m to 0.24-0.26 m, with the largest improvement in non-line-of-sight snapshots. The multi-cell non-coherent detector further lowers the computational complexity without degrading positioning accuracy, demonstrating its suitability for practical multipath radio-SLAM.
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization