When Does Inter-BS Interference Hurt Sensing in ISAC? An Exact Fisher Information Analysis under Finite-Alphabet Interference
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: I'm Rosa.
Dev: And I'm Dev, here with Taro.
Rosa: We're looking at the paper: "When Does Inter-BS Interference Hurt Sensing in ISAC? An Exact Fisher Information Analysis under Finite-Alphabet Interference".
Dev: Here's the gist of the paper. In networked integrated sensing and communication (ISAC), a base station performing sensing using its own downlink signals is subject to interference from neighboring BSs' downlink transmissions This paper investigates the impact of the signal constellation and waveform on the sensing performance in the presence of such inter-BS interference.
Rosa: Impact on Linear Receivers For any linear receiver, we show that the mean-squared crossambiguity between two random ISAC signals is invariant to the modulation bases and constellations We then derive the exact Fisher information under finite-alphabet interference with a flat-fading link, which equals the interference-free Fisher information scaled by 1 − γ MMSES (γ), where γ denotes the interference-to-noise ratio (INR) and MMSES is the minimum mean-square error (MMSE) in estimating an interfering data symbol from its noise-corrupted observation at the sensing receiver.
Dev: Consequently, Gaussian interference is the worst case and corresponds to the CRB loss of the matched filtering, whereas finite-alphabet interference induces a non-monotonic CRB loss that peaks at intermediate INR and vanishes at high INR.
Rosa: The loss is waveform-independent for the flat-fading BS-BS link, and an expectation-maximization (EM) interference-cancellation receiver approaches the theoretical bound for delay estimation.
Paper summary: Rosa: We’re looking at this paper now, "When Does Inter-BS Interference Hurt Sensing in ISAC? An Exact Fisher Information Analysis under Finite-Alphabet Interference." Essentially, the core thesis is about how signal constellation and waveform choice impacts sensing performance when you have interference from other base stations.
Dev: They start by showing that for any linear receiver, the mean squared cross ambiguity between two random ISAC signals is actually invariant to the modulation bases and constellations. That’s a strong starting point because it means the fundamental ambiguity doesn't depend on how you encode your data, which simplifies things initially.
Taro: But they don't stop there; they then derive the exact Fisher information under finite-alphabet interference with a flat fading link. This derivation is what connects all those constellation choices and waveforms to the final sensing accuracy metric.
Rosa: The key finding here is that this derived Fisher information equals the interference-free Fisher information scaled by one minus gamma MMSES of gamma, where gamma stands for the interference to noise ratio. This directly tells us how much performance scales with that ratio.
Dev: And they pinpoint Gaussian interference as the worst case, which corresponds to the constant rate bound loss of matched filtering. But they show that finite-alphabet interference results in a nonmonotonic constant rate bound loss that peaks at an intermediate gamma and then goes to zero at high gamma.
Taro: That nonmonotonic behavior is what really makes it interesting for autonomy research because it suggests there's a sweet spot where interference isn't totally detrimental, unlike just assuming more power always means worse results.
Rosa: And they also state that the loss is waveform independent for flat-fading BS to BS links, and an expectation maximization based soft interference cancellation receiver approaches the theoretical bound for delay estimation under these conditions.
Dev: That suggests that if we use a proper soft cancellation receiver, we can actually reach the theoretical limit even when dealing with this finite-alphabet interference. So, the paper is arguing that it’s about how you structure your cancellation and your system design to handle the noise correctly.
Conclusion: Rosa: So, looking at the whole thing, the main message of this work by Kawon Han, Kaitao Meng, Fan Liu, and Christos Masouros is that for linear receivers in ISAC sensing with inter-BS interference, the impact on performance doesn't depend on your specific modulation or waveform.
Dev: They established that this dependence is handled because the exact Fisher information gets scaled by one minus gamma MMSES of gamma, which means the loss peaks at some intermediate interference to noise ratio and then drops off completely at high INR.
Taro: What this means for us in terms of autonomy and deployment is that we should try to avoid those intermediate interference points entirely because increasing the interference power there might actually make the sensing worse.
Rosa: Right, so it’s about designing systems that operate outside those specific regions where adding more interference power would hurt performance, which is a practical constraint for deploying this technology.
Dev: And they showed that if you use an expectation maximization based soft cancellation receiver, you can actually get close to the theoretical limit even when the interference is finite-alphabet. That points toward a robust way to handle these complex real-world scenarios without needing perfect knowledge of every detail.
University of Nanistructure and Information Science (UNIST) · The University of Manchester · Southeast University · University College London
eess.SP
Submitted: 2026-11-05
Updated: 2026-11-05
Comments: 6 pages, 4 figures, submitted to IEEE possible publication
License: http://creativecommons.org/publicdomain/zero/1.0/
Importance score: 83/100
The gist: In networked integrated sensing and communication (ISAC), a base station performing sensing using its own downlink signals is subject to interference from neighboring BSs' downlink transmissions This…
Key concepts
- Fisher Information Matrix (FIM)
- The FIM quantifies the amount of information available about the signal parameters being estimated. In this paper, it is derived exactly for sensing under finite-alphabet interference, showing how the interference modifies this information and thus affects estimation accuracy.
- CRB Loss ($L_S(\gamma)$)
- This measures the degradation in performance due to interference, defined as $1 - \gamma$ MMSES($\gamma$). It shows that this loss is not monotonic; it peaks at an intermediate interference-to-noise ratio ($\gamma^*$) and disappears when the interference power is very high.
- Finite-Alphabet Interference
- This refers to interference caused by a limited set of possible signal constellations (like QAM or PSK) rather than continuous Gaussian noise. The analysis shows that this type of interference leads to a non-monotonic loss curve, unlike the worst-case Gaussian interference which always degrades performance.
- Waveform Independence
- The paper demonstrates that for flat-fading links, the sensing performance loss is independent of the specific waveform used by the interferer (e.g., OFDM or OTFS). This means different communication schemes do not inherently make sensing more or less vulnerable to inter-BS interference under these conditions.
Terminology
Summary
In networked integrated sensing and communication (ISAC), a base station performing sensing using its own downlink signals is subject to interference from neighboring BSs' downlink transmissions This paper investigates the impact of the signal constellation and waveform on the sensing performance in the presence of such inter-BS interference.
Impact on Linear Receivers
For any linear receiver, we show that the mean-squared crossambiguity between two random ISAC signals is invariant to the modulation bases and constellations We then derive the exact Fisher information under finite-alphabet interference with a flat-fading link, which equals the interference-free Fisher information scaled by 1 − γ MMSES (γ), where γ denotes the interference-to-noise ratio (INR) and MMSES is the minimum mean-square error (MMSE) in estimating an interfering data symbol from its noise-corrupted observation at the sensing receiver. Consequently, Gaussian interference is the worst case and corresponds to the CRB loss of the matched filtering, whereas finite-alphabet interference induces a non-monotonic CRB loss that peaks at intermediate INR and vanishes at high INR. p. 0 We then derive the
exact Fisher information under finite-alphabet interference with
a flat-fading link, which equals the interference-free Fisher
information scaled by 1 − γ MMSES (γ), where γ denotes the
interference-to-noise ratio (INR) and MMSES is the minimum
mean-square error (MMSE) in estimating an interfering data
symbol from its noise-corrupted observation at the sensing
receiver. Consequently, Gaussian interference is the worst case
and corresponds to the CRB loss of the matched filtering, whereas
finite-alphabet interference induces a non-monotonic CRB loss
that peaks at intermediate INR and vanishes at high INR. Consequently, Gaussian interference is the worst case
and corresponds to the CRB loss of the matched filtering, whereas
finite-alphabet interference induces a non-monotonic CRB loss
that peaks at intermediate INR and vanishes at high INR. The
loss is waveform-independent for the flat-fading BS-BS link,
and an expectation-maximization (EM) interference-cancellation
receiver approaches the theoretical bound for delay estimation.Where the paper says it
Where the paper says it
Fisher Information Derivation
The exact Fisher information matrix (FIM) is derived under finite-alphabet interference via a complex Brown identity, revealing that Gaussian interference is the worst case and that finite-alphabet interference yields a nonmonotonic loss that vanishes at high INR. The CRB loss LS (γ) is defined as 1 − γ MMSES (γ). p. 0 We then derive the
exact Fisher information under finite-alphabet interference with
a flat-fading link, which equals the interference-free Fisher
information scaled by 1 − γ MMSES (γ), where γ denotes the
interference-to-noise ratio (INR) and MMSES is the minimum
mean-square error (MMSE) in estimating an interfering data
symbol from its noise-corrupted observation at the sensing
receiver. Consequently, Gaussian interference is the worst case
and corresponds to the CRB loss of the matched filtering, whereas
finite-alphabet interference induces a non-monotonic CRB loss
that peaks at intermediate INR and vanishes at high INR.Where the paper says it
((i)) 1 ≤ LS (γ) ≤ 1 + γ, where the upper bound holds with equality for Gaussian S1 and equals the loss of MF, i.e., treating interference as noise (TIN). p. 2 The CRB loss LS (γ) satisfies the following
properties:
(i) 1 ≤ LS (γ) ≤ 1 + γ, where the upper bound holds with
equality for Gaussian S1 and equals the loss of MF, i. ,
treating interference as noise (TIN). Substituting these bounds into (11) yields 1 ≤
LS (γ) ≤ 1 + γ. The upper bound is attained for Gaussian S1,
for which the conditional mean estimator is linear.Where the paper says it
Where the paper says it
((ii)) limγ→0 LS (γ)−1 γ= 1. p. 2 ,
treating interference as noise (TIN). (ii) limγ→0
LS (γ)−1
γ
= 1. (ii) limγ→0
LS (γ)−1
γ
= 1. Hence, at low INR, finite-alphabet
interference is as harmful as Gaussian interference.Where the paper says it
Where the paper says it
((iii)) LS (γ) → 1 as γ → ∞ for any finite S. p. 2 Hence, at low INR, finite-alphabet
interference is as harmful as Gaussian interference. (iii) LS (γ) → 1 as γ → ∞ for any finite S.Where the paper says it
((iv)) LS (γ) is maximized at a finite INR γ⋆. p. 2 (iii) LS (γ) → 1 as γ → ∞ for any finite S. (iv) LS (γ) is maximized at a finite INR γ
⋆
.Where the paper says it
Waveform and Channel Dependence
The waveform of the interferer affects the FIM only through A = HIU1, and Corollary 1 yields a closed form when A has orthogonal columns. Proposition 3 provides a sufficient condition for waveform independence: if the BS-BS channel is flat-fading, i. p. 2 Dependence of the FIM on the Interferer Waveform
By Theorem 1, the waveform of the interferer affects the
FIM only through A = HIU1, and Corollary 1 yields a closed
form when A has orthogonal columns. The following proposition gives a sufficient condition for waveform independence. If the BS-BS channel is flat-fading, i. ,
HI = hI, then Jw = ρS (γ)I/σ2
for every unitary U1. ,
HI = hI, then Jw = ρS (γ)I/σ2
for every unitary U1. Hence, LS (γ) is identical for every interferer waveform (SC,
CP-OFDM, OTFS, DFT-s-OFDM, AFDM, etc. ■
Note that F is conditioned on the known s0 and marginalized over s1. Since F = ρS (γ)F0 holds for every realization
of s0, LS (γ) is identical whether or not the CRB is further
averaged over s0. Thus, Jw =
ρS (γ)I/σ2 does not depend on U1, and F = ρS (γ)F0 follows
for any µ(η), i. ■
Under a frequency-selective channel, HI =
WH
N diag(Hk)WN , so that A has orthogonal columns
for CP-OFDM (Q = WH
N , γk = |Hk|
2/σ2
), whose loss
averages the scalar losses over the channel-induced INR
spread.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
Receiver Performance and Cancellation
An expectation-maximization (EM)-based soft interference-cancellation receiver approaches the theoretical bound for delay estimation. The EM receiver runs 10 iterations with a FFT-based delay estimator in the M-step, and the RMSE is averaged over 1000 trials. p. 3 a per-sample echo SNR of −10 dB and BS1 transmits 16QAM
or 64QAM data. The EM receiver runs 10 iterations with a
FFT-based delay estimator in the M-step, and the RMSE is
averaged over 1000 trials. The EM receiver runs 10 iterations with a
FFT-based delay estimator in the M-step, and the RMSE is
averaged over 1000 trials. Since Corollary 1 scales the entire
FIM, the delay CRB is inflated by the same factor LS (γ). The RMSE of MF follows the TIN bound, which equals
√
1 + γ times the interference-free bound, and breaks down
at γ ≈ 10 dB, where the integrated SINR falls below the
detection threshold. In contrast, the EM receiver follows the
finite-alphabet bound and enables sensing at INRs where TIN
fails completely.Where the paper says it
Where the paper says it
Where the paper says it
Numerical Evaluation
The CRB loss LS (γ) in (11) versus INR is evaluated for various constellations. All constellations follow the Gaussian curve at low INR, reach a peak, and then collapse to 0 dB. p. 3 2 evaluates the CRB loss in (11) for various constellations. All constellations follow the Gaussian curve at low
INR, reach a peak, and then collapse to 0 dB. (dB)
-10 0 10 20 30 40 50 60
CRB loss, L
S
(. ) (dB)
0
5
10
15
20
25
TIN
QPSK
16-QAM
64-QAM
256-QAM
1024-QAM
8-PSK
16-PSK
Figure 2. 5 dB, respectively. That is, each fourfold increase of
the QAM order raises the peak by approximately 2. That is, each fourfold increase of
the QAM order raises the peak by approximately 2. 9 dB and
shifts it by approximately 6 dB. 5 dB owing to its constant amplitude. For
64QAM at γ = 20 dB, MF suffers a 20 dB CRB loss, whereas
the optimal receiver can lose only 2 dB.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
Conclusion
In this paper, we have presented an estimation-theoretic framework for ISAC sensing under finite-alphabet inter-BS interference. For linear receivers, the impact of the inter-BS interference is invariant to the constellation and waveform. p. 3 CONCLUSION
In this paper, we have presented an estimation-theoretic
framework for ISAC sensing under finite-alphabet inter-BS
interference. For linear receivers, the impact of the inter-BS
interference is invariant to the constellation and waveform. Average Fisher-information loss for a QPSK interferer over (a) a flat-fading channel, (b) a two-tap channel, and (c) a three-tap channel. peaks at intermediate INR, vanishes at high INR, is waveformindependent for flat-fading and synchronized BS-BS links, and
is approached by EM-based soft cancellation. peaks at intermediate INR, vanishes at high INR, is waveformindependent for flat-fading and synchronized BS-BS links, and
is approached by EM-based soft cancellation. Hence, inter-BS
coordination should avoid intermediate INRs, where increasing
the interference power may even improve sensing.Where the paper says it
Where the paper says it
Where the paper says it
APPENDIX A
Since E[s1s H 1] = IN and G is unitary, the second moment is E[χ 2] = N. For the fourth moment, let g = Gs0 so that χ = P i g s 1,i. p. 4 APPENDIX A
PROOF OF PROPOSITION 1
Since E[s1s
H
1
] = IN and G is unitary, the second moment
is
E[|χ|
2
] = E
s
H
0 GHE[s1s
H
1
]Gs0
= E
∥s0∥
2
= N. (13)
For the fourth moment, let g = Gs0 so that χ =
P
i
gis
∗
1,i. (13)
For the fourth moment, let g = Gs0 so that χ =
P
i
gis
∗
1,i. proper symbols with unit power and kurtosis µ4, the
only non-zero fourth-order moments are E[s
∗
i
s
∗
j
sksl
] = µ4 for
i = j = k = l and 1 for {i, j} = {k, l} with i ̸= j, since
E[s
2
] = 0. In a dense, TDD-synchronized network, a BS that performs full-duplex monostatic sensing inevitably receives the
DL transmissions of its neighbors on the same time-frequency
resources [3], [4]. Since BS-to-BS links are typically line-ofsight (LoS), this inter-BS interference can exceed the thermal
noise by tens of decibels, and it fundamentally limits the
sensing performance of network-level ISAC [5].Where the paper says it
Where the paper says it
Where the paper says it
APPENDIX B
The Wirtinger chain rule with ∂(jw)∗/∂w∗ = -jI yields E[sws T w] = 0. This implies that E[sws s T w] = 0. p. 4 Under (A1) and (A4), the circularity of n implies that jw
has the same distribution as w, satisfying pw(jw) = pw(w). The Wirtinger chain rule with ∂(jw)
∗/∂w∗ = −jI yields
sw(jw) = jsw(w), and therefore
E[sws
T
w] = E
sw(jw)sw(jw)
T
= −E[sws
T
w], (19)
which implies E[sws
T
w] = 0.Where the paper says it
APPENDIX C
The log-likelihood is given by ln pw(r − µ(η)). The chain rule gives ∂/∂ηi ln pw(r − µ) = -s H w a i - s T w a i = -2R[s H w a i]. p. 4 Let ai =
∂µ/∂ηi
. Since ln pw is real-valued, ∇w ln pw = s
∗
w, and the
chain rule gives
∂
∂ηi
ln pw(r − µ) = −s
H
wai − s
T
wa
∗
i = −2ℜ{s
H
wai}.Where the paper says it
APPENDIX D
If A = QΛ, the sufficient statistic v = QHw consists of N independent scalar channels vi = λi s 1,i + ˜ni, since QHn ∼ CN (0, σ2I) and the entries of s1 are i.i.d. Normalizing by σ and noting that the MMSE is invariant to the known phase of λi, we have E[cov(s1w)] = diag(MMSES (γi)), and thus E[cov(As1w)] = σ2Q diagγi MMSES (γi) QH. p. 5 (22)
APPENDIX D
PROOF OF COROLLARY 1
If A = QΛ, the sufficient statistic v = QHw =
Λs1 + QHn consists of N independent scalar channels
vi = λis1,i + ˜ni
, since QHn ∼ CN (0, σ2I) and the
entries of s1 are i. Normalizing by σ and noting that
the MMSE is invariant to the known phase of λi
, we have
E[cov(s1|w)] = diag(MMSES (γi)), and thus
E
cov(As1|w)
= QΛ diagMMSES (γi)
Λ
HQH
= σ
2Q diagγi MMSES (γi)
QH.Where the paper says it
APPENDIX E
The upper bound is attained for Gaussian S1, for which the conditional mean estimator is linear. From (i), we have 0 ≤ MMSES (γ) ≤ 1/(1 + γ). p. 0 For
any linear receiver, we show that the mean-squared crossambiguity between two random ISAC signals is invariant to
the modulation bases and constellations. We then derive the
exact Fisher information under finite-alphabet interference with
a flat-fading link, which equals the interference-free Fisher
information scaled by 1 − γ MMSES (γ), where γ denotes the
interference-to-noise ratio (INR) and MMSES is the minimum
mean-square error (MMSE) in estimating an interfering data
symbol from its noise-corrupted observation at the sensing
receiver. (24)
As γ → 0, the observation Y =
√γS1 +N becomes independent of S1. Consequently, E[S1|Y ] converges to E[S1] = 0,
implying MMSES (γ) → E[|S1|
2
] = 1. Consequently, E[S1|Y ] converges to E[S1] = 0,
implying MMSES (γ) → E[|S1|
2
] = 1. Since γMMSES (γ) →
0, it follows that limγ→0(LS (γ) − 1)/γ = 1. (26)
Combining these inequalities yields
0 ≤ γMMSES (γ) ≤
(M − 1)d
2
max
2
γe−γd2
min/4
. (27)
Since dmin > 0, the right-hand side converges to zero as γ →
∞. Therefore, there exist 0 <
a < b < ∞ such that the global maximum is attained within
[a, b]. By continuity, LS (γ) attains its maximum at some finite
γ
⋆ ∈ (0,∞), completing the proof.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
REFERENCES
[1] F. Liu, Y.-F. Liu, Y. Cui, C. Masouros, J. p. 0 uk
Fan Liu
Southeast University
Nanjing, China
fan. cn,
Christos Masouros
University College London
London, United Kingdom
c.Where the paper says it
[2] C Sturm and W Wiesbeck
[3] K Han, K Meng, X.-Y Wang, and C Masouros
[4] K Meng, K Han, C Masouros, and L Hanzo
[5] K Meng, C Masouros, G Chen, and F Liu
[6] F Liu, Y Zhang, Y Xiong, S Li, W Yuan, F Gao
[7] M F Keskin et al
[8] Z Du et al
[9] B Geiger et al
[10] K Meng, K Han, C Masouros, and F Liu
[11] K Han, K Meng, A Chatzicharistou, and C Masouros
[12] K Meng, K Han, C Masouros, and L Hanzo
[13] H Pu et al
[14] A Dytso et al
[15] P Stoica and P Babu
[16] L D Brown
[17] M Payaro and D Palomar
[18] A P Dempster et al
[19] X Wang and H V Poor
[20] S M Kay Fundamentals of Statistical Signal Processing: Estimation Theory. Upper Saddle River, NJ: Prentice Hall, 1993
[20] S M Kay Fundamentals of Statistical Signal Processing: Estimation Theory.
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
-
Ability to achieve sensing performance independent of constellation design for linear receivers by exploiting that
the mean-squared crossambiguity between two random ISAC signals is invariant to the modulation bases and constellations.
This means a linear receiver does not need complex knowledge of the interferer's constellation or waveform. -
Capability to precisely characterize sensing accuracy under finite-alphabet interference by using
the exact Fisher information matrix (FIM) under finite-alphabet interference,
which allows for a more accurate prediction of delay estimation than methods based on Gaussian surrogates. -
Implementation of an Expectation-Maximization (EM) based soft interference cancellation receiver that
approaches the theoretical bound for delay estimation,
enabling sensing at InRs where traditional matched filtering fails completely. -
Design of receivers that exploit the non-monotonic CRB loss, which
peaks at intermediate INR and vanishes at high INR,
guiding network coordination to avoid specific interference power levels for optimal performance. -
Waveform-independent performance guarantees for flat-fading BS-BS links, meaning the sensing accuracy
is waveform-independent for the flat-fading BS-BS link,
simplifying receiver design across different OFDM variants like SC, CP-OFDM, and OTFS.
Abstract
In networked integrated sensing and communication (ISAC), a base station (BS) performing sensing using its own downlink signals is subject to interference from the downlink transmissions of neighboring BSs. This paper investigates the impact of the signal constellation and waveform on the sensing performance in the presence of such inter-BS interference. For any linear receiver, we show that the mean-squared cross-ambiguity between two random ISAC signals is invariant to the modulation bases and constellations. We then derive the exact Fisher information under finite-alphabet interference with a flat-fading link, which equals the interference-free Fisher information scaled by 1-γ, (γ), where γ denotes the interference-to-noise ratio (INR) and is the minimum mean-square error (MMSE) in estimating an interfering data symbol from its noise-corrupted observation at the sensing receiver. Consequently, Gaussian interference is the worst case and corresponds to the CRB loss of the matched filtering, whereas finite-alphabet interference induces a non-monotonic CRB loss that peaks at intermediate INR and vanishes at high INR. The loss is waveform-independent for the flat-fading BS-BS link, and an expectation-maximization (EM) interference-cancellation receiver approaches the theoretical bound for delay estimation.
Sources
- Constellation Selection and Power Allocation for Multi-Cell OFDM-ISAC: Managing Inter-Cell Interference and Sensing Sidelobes
- Blind OFDM-ISAC Relying on Asymmetric Modem Constellations
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