Multi-Channel Secure Communication (MCSC)

• Multi-Channel Secure Communication (MCSC) is a framework that leverages diverse channel resources, such as multiple antennas, relay links, and heterogeneous media, to strengthen security through specialized coding and beamforming.
• It encompasses a range of models—from spatial wiretap and multiaccess schemes to robust transceiver designs—addressing trade-offs in secrecy capacity based on channel geometry and eavesdropper knowledge.
• MCSC principles extend to practical applications in THz systems, IoT networks, and vehicular communications by integrating advanced encryption, noise injection, and dynamic channel management.

Multi-Channel Secure Communication (MCSC) denotes, in the cited literature, a family of secure-communication models and systems in which security is obtained by exploiting multiplicity in the communication substrate: multiple antennas and spatial modes, multiple-access links, parallel relay subchannels, multiple coordinated base stations, multiple physical or logical channels, or heterogeneous channels such as wireless and optical links. Across these works, the governing objectives include secrecy capacity under $I(W;Y_e^n)/n\to0$, rate-equivocation, minimum MSE constraints at eavesdroppers, simulation-based correctness and privacy under mixed channel corruption, secure-key rates in quantum multiple-access channels, and authentication accuracy in dual-channel challenge-response systems (0708.4219, Khisti et al., 2010, Jagyasi et al., 2020, Vaya, 2010, Das et al., 2021, Vincenzi et al., 1 May 2025).

1. Scope and recurring problem formulations

The literature does not impose a single canonical formalization of MCSC. In the MISOME wiretap model, the transmitter has $n_t$ antennas, the intended receiver has a single antenna, the eavesdropper has $n_e$ antennas, the channels are fixed and known to all terminals, and the secrecy requirement is to design an $(n,2^{nR})$ code such that the probability of decoding error at the receiver tends to $0$ and $I(W;Y_e^n)/n\to0$ (0708.4219). In the MIMOME model, the sender, intended receiver, and eavesdropper all have multiple antennas, and secrecy capacity is written as a saddle point solution to a minimax problem (Khisti et al., 2010). In the MAWC-CM model, two transmitters have confidential messages, both have access to a common message, and the eavesdropper must decode only the common message (Zivari-Fard et al., 2014). In the parallel relay-eavesdropper model, secure communication is realized over multiple relay-eavesdropper channels viewed as subchannels (Awan et al., 2010). In robust multicast MCC, the objective is to minimize the Sum-Mean-Square-Error at legitimate users while enforcing $\epsilon_e\ge\Gamma$ for each eavesdropper (Jagyasi et al., 2020).

At the protocol level, MCSC also appears in models in which channels themselves are corruption objects. In unconditional secure multiparty computation with man-in-the-middle attacks, each directed channel is secure, authenticated-but-eavesdroppable, partially-tamperable, or fully-tamperable, and the security definitions distinguish honest parties that retain correctness from those that retain privacy (Vaya, 2010). In quantum communication, a generalized quantum multiple-access channel is the composition of a forward CP TP map, local encoding CPTP maps at two senders, and a backward CP TP map (Das et al., 2021). In application-oriented systems, MCSC is implemented through frequency-multiplexed OOK links, polarization channels, multiple insecure steganographic channels, dynamic channel hopping in IoT, or coupled NLOS and LOS channels in vehicular authentication (Farzin et al., 2024, Farzin et al., 2024, Omego et al., 8 Jan 2025, Barman et al., 11 Sep 2025, Vincenzi et al., 1 May 2025).

Formulation	Channel multiplicity	Security objective
MISOME / MIMOME wiretap	antennas, eigenmodes, GSVD modes	secrecy capacity
MAWC-CM / cooperating encoders / relay-eavesdropper	users, conferencing links, subchannels, relay modes	rate-equivocation / perfect secrecy
Mixed-corruption MPC / quantum GMAC	pairwise channels, forward-backward quantum links	correctness and privacy / secret-key rates
THz metasurfaces / steganography / IoT / vehicular	frequency channels, polarization channels, logical channels, channel hopping, LOS+NLOS	BER, MAC freshness, resilience, authentication accuracy

This diversity suggests that “channel” in MCSC is not restricted to frequency partitioning. It can denote spatial degrees of freedom, relay branches, cooperative encoders, logical cover channels, or heterogeneous physical media.

2. Spatial wiretap channels and secrecy capacity

The classical information-theoretic core of MCSC is the multi-antenna wiretap problem. For MISOME, with $h\in\mathbb C^{n_t\times 1}$ and $G\in\mathbb C^{n_e\times n_t}$, the secrecy capacity is

$$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$

The optimum covariance is rank one, $n_t$0, where $n_t$1 is the principal generalized eigenvector of $n_t$2, and the secrecy capacity admits the closed form

$n_t$3

Accordingly, a beamforming strategy is capacity-achieving, and the operational interpretation is to concentrate all power along the spatial direction that maximizes the ratio of SNR at the receiver to SNR at the eavesdropper (0708.4219).

The high-SNR structure is sharply geometry-dependent. If $n_t$4 lies partly in $n_t$5, then at $n_t$6,

$n_t$7

so the secrecy capacity grows unbounded. If $n_t$8, then

$n_t$9

and capacity saturates. The same work introduces masked beamforming,

$n_e$0

with $n_e$1 isotropic in the orthogonal subspace, and shows

$n_e$2

so that, up to an SNR penalty of $n_e$3 $n_e$4, there is no gain from knowing $n_e$5 at high SNR. In the large-antenna regime, with $n_e$6, the high-SNR limit satisfies

$n_e$7

so that for $n_e$8 the eavesdropper can drive the secrecy capacity to zero (0708.4219).

For MIMOME, the secrecy-capacity characterization is broader and more algebraic. With Bob’s channel $n_e$9 and Eve’s channel $(n,2^{nR})$0, the secrecy capacity is

$(n,2^{nR})$1

where $(n,2^{nR})$2 is a noise-cross-covariance matrix and $(n,2^{nR})$3 is the input covariance. At the saddle point $(n,2^{nR})$4,

$(n,2^{nR})$5

At high SNR, the channel pair $(n,2^{nR})$6 is simultaneously diagonalized by the generalized singular value decomposition, producing parallel subchannels with generalized singular values $(n,2^{nR})$7, and the secrecy rate becomes

$(n,2^{nR})$8

In the very high-SNR regime one uses only modes with $(n,2^{nR})$9, and

$0$0

The zero-capacity condition is exact:

$0$1

In the many-antenna limit, the scaling laws establish that to prevent secure communication, the eavesdropper needs $0$2 times as many antennas as the sender and intended receiver have jointly, and that the optimum division of antennas between sender and intended receiver is in the ratio of $0$3. A central caveat is that semi-blind “masked” MIMO can be arbitrarily far from capacity in this regime (Khisti et al., 2010).

Taken together, these results make the spatial version of MCSC a problem of generalized eigenvalue selection, GSVD mode activation, and power allocation on secure modes rather than a problem of merely adding antennas.

3. Multi-user, relay, and multiplex coding models

In multiple-access secrecy problems, the multiplicity of channels is often user-induced rather than spatial. The MAWC-CM model consists of two transmitters with confidential messages $0$4 and $0$5, a common message $0$6 decoded by both receivers, and secrecy requirement

$0$7

The discrete-memoryless outer bound uses auxiliary variables $0$8, while the inner bound generates a cloud-center codebook $0$9 for the common message and two private subcodebooks that are randomly binned. The role of the auxiliaries is explicit: $I(W;Y_e^n)/n\to0$0 carries common message, decodable at both; $I(W;Y_e^n)/n\to0$1 superimpose private layers on $I(W;Y_e^n)/n\to0$2; protected by random-binning against $I(W;Y_e^n)/n\to0$3. For the switch-channel special case, the inner and outer bounds coincide. In the Gaussian version, superposition parameters $I(W;Y_e^n)/n\to0$4 trade common-layer rate against private secrecy margin, and numerical examples show that when the eavesdropper’s channel is much noisier, the secrecy-constraint region can exceed the compound-MAC region (Zivari-Fard et al., 2014).

A closely related variant is the multiaccess channel with partially cooperating encoders and security constraints. Encoder $I(W;Y_e^n)/n\to0$5 holds the confidential message $I(W;Y_e^n)/n\to0$6, Encoder $I(W;Y_e^n)/n\to0$7 has no message of its own, and a unidirectional noiseless bit-pipe of capacity $I(W;Y_e^n)/n\to0$8 enables conferencing. The inner bound is based on a combination of Willems's coding scheme, noise injection and additional binning that provides randomization for security. In the Gaussian model, $I(W;Y_e^n)/n\to0$9 reduces to a wiretap channel with a helper interferer, whereas $\epsilon_e\ge\Gamma$0 yields the two-antenna transmitter wiretap channel. The numerical examples identify a geometric operating rule: when the helper is near the legitimate receiver, its best strategy is to inject noise; when it is nearer to Encoder $\epsilon_e\ge\Gamma$1, it prefers to forward message through conferencing (Awan et al., 2012).

In relay-assisted MCSC, the decisive degree of freedom is subchannel-by-subchannel mode selection. The parallel relay-eavesdropper channel consists of $\epsilon_e\ge\Gamma$2 relay-eavesdropper subchannels. The inner bound allows mode selection at the relay: on each subchannel the relay either decodes-and-forwards the source message or confuses the eavesdropper through noise injection. In the Gaussian memoryless model, the achievable secrecy rate is obtained by optimizing powers $\epsilon_e\ge\Gamma$3 and correlation parameters on the DF subchannels, while the upper bound is a per-subchannel sum of secrecy differences. In the “deaf relay” special case, the lower and upper bounds coincide under a stated condition. Numerical examples with $\epsilon_e\ge\Gamma$4 subchannels show that optimized power allocation yields up to $\epsilon_e\ge\Gamma$5 gain in secrecy rate, and that per-tone mode selection outperforms fixed DF-only or NF-only policies in the intermediate relay-source-distance regime (Awan et al., 2010).

A distinct but adjacent coding development is secure multiplex coding with dependent and non-uniform multiple messages. Its purpose is to remove rate loss in the coding for wire-tap channels and broadcast channels with confidential messages caused by the inclusion of random bits into transmitted signals: secure multiplex coding replaces the random bits by other meaningful secret messages. The generalization of channel resolvability yields leakage bounds for dependent and non-uniform message collections, and under SACU one also shows strong secrecy (Hayashi et al., 2012).

These multi-terminal models broaden MCSC beyond spatial beamforming. They formalize secrecy through superposition, binning, conferencing, friendly jamming, relay mode selection, and the use of multiple meaningful secret messages as intrinsic randomization.

4. Robust multi-base-station transceiver design

A more engineering-oriented branch of MCSC treats secure communication as a joint beamforming, artificial-noise, and robustness problem. In the downlink MIMO-multicast cluster considered for mission-critical communications, $\epsilon_e\ge\Gamma$6 coordinated base stations, each with $\epsilon_e\ge\Gamma$7 antennas, serve a common message $\epsilon_e\ge\Gamma$8 to $\epsilon_e\ge\Gamma$9 legitimate users, each with $h\in\mathbb C^{n_t\times 1}$0 antennas, in the presence of $h\in\mathbb C^{n_t\times 1}$1 passive eavesdroppers, each with $h\in\mathbb C^{n_t\times 1}$2 antennas. The transmit signal at BS $h\in\mathbb C^{n_t\times 1}$3 is

$h\in\mathbb C^{n_t\times 1}$4

and the per-BS power is constrained by

$h\in\mathbb C^{n_t\times 1}$5

The joint design minimizes the Sum-MSE at the legitimate users,

$h\in\mathbb C^{n_t\times 1}$6

subject to $h\in\mathbb C^{n_t\times 1}$7 for all eavesdroppers and the per-BS power constraints (Jagyasi et al., 2020).

The same framework studies two CSI-error models. Under stochastic Gaussian errors, the MSE expressions acquire extra terms such as

$h\in\mathbb C^{n_t\times 1}$8

and the problem is nonconvex jointly but convex in each block $h\in\mathbb C^{n_t\times 1}$9. This motivates Algorithm 1, a coordinate-descent scheme in which eavesdropper filters are updated in MMSE form, legitimate receive filters satisfy $G\in\mathbb C^{n_e\times n_t}$0, Lagrange multipliers enforce $G\in\mathbb C^{n_e\times n_t}$1 and power constraints, precoders satisfy $G\in\mathbb C^{n_e\times n_t}$2, and AN shaping matrices $G\in\mathbb C^{n_e\times n_t}$3 are taken as normalized null-space bases of effective matrices $G\in\mathbb C^{n_e\times n_t}$4. Under norm-bounded errors, $G\in\mathbb C^{n_e\times n_t}$5 and $G\in\mathbb C^{n_e\times n_t}$6, the design becomes a three-stage worst-case iterative algorithm (Jagyasi et al., 2020).

Several concrete numerical findings are part of the model’s significance. Complexity per iteration is dominated by $G\in\mathbb C^{n_e\times n_t}$7 matrix inversions of size up to $G\in\mathbb C^{n_e\times n_t}$8, i.e. $G\in\mathbb C^{n_e\times n_t}$9, and empirically $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$0 coordinate iterations suffice, while the worst-case outer loop needs $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$1 iterations. At BER $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$2, robust design under stochastic errors shows up to a $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$3 dB BER gain, and NBE-robust design gives $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$4 dB. Without AN the security gap at target BER $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$5 is $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$6 dB; with NBE-robust + AN it drops to $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$7 dB; with SE-robust + AN even to $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$8 dB. Adding AN slightly worsens legitimate BER but dramatically raises eavesdropper BER/MSE, ensuring $$C_s = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ I(x; y_r) - I(x; y_e ) ]^+ = \max_{Q\succeq0,\ \mathrm{tr}(Q)\le P} [ \log \det(1 + h^H Q h ) - \log \det( I + G Q G^H ) ]^+.$$9 over a wide SNR range. At system level, MBSFN yields the best user-BER CDF and the largest legitimate-vs-eavesdropper BER gap, whereas dynamic clustering trades off capacity versus reliability/security (Jagyasi et al., 2020).

This branch of MCSC is not a secrecy-capacity theory in the strict Shannon sense. It is a robust transceiver synthesis framework in which beamforming, AN injection, and CSI uncertainty are treated jointly, and security is operationalized through guaranteed eavesdropper MSE and security-gap behavior.

5. Unconditional and quantum formulations

MCSC also appears in models where the central issue is not Gaussian signaling but the structure of the communication network and the security definition itself. In unconditional secure multiparty computation with man-in-the-middle attacks, every directed channel $n_t$00 is one of four types: secure, authenticated-but-eavesdroppable, partially-tamperable, or fully-tamperable. The corruption pattern is collected in a sextuplet

$n_t$01

For a given corruption $n_t$02, the unsacrificed sets are defined through cliques of honest parties: $n_t$03 is the largest subset of honest parties that form a clique using only secure or eavesdroppable channels and has size $n_t$04, while $n_t$05 is the largest subset that uses only secure channels in a clique of size $n_t$06. The protocol is required to satisfy correctness for all parties in $n_t$07 and privacy for all parties in $n_t$08, with simulation-based privacy defined by a simulator that receives the adversary’s code, the inputs and committed values of the sacrificed or corrupted parties, and the final output (Vaya, 2010).

To realize these guarantees over mixed-corruption channels, the construction adapts any information-theoretically secure $n_t$09-party protocol $n_t$10 by slowing it down by a factor of $n_t$11 and assigning one super-round per ordered pair per original round. Theorem 4.1 states that for any feasible adversary structure that corrupts at most $n_t$12 parties and arbitrary subsets of channels, the resulting protocol securely evaluates any $n_t$13. Theorem 4.2 gives the threshold’s optimality: no information-theoretic protocol can tolerate $n_t$14 Byzantine parties and arbitrary channel corruptions and still achieve both correctness and privacy for any non-trivial $n_t$15 (Vaya, 2010).

The quantum version of MCSC takes a different form. In the generalized quantum MAC, Alice prepares a GHZ state

$n_t$16

sends the flying qubits through a forward CP TP map to Bob$n_t$17 and Bob$n_t$18, receives encoded qubits back through a backward CP TP map, and decodes using one of two GHZ bases. Bob$n_t$19 applies one of $n_t$20 to encode two key bits $n_t$21; Bob$n_t$22 applies $n_t$23 to encode a secret bit $n_t$24 and an auxiliary bit $n_t$25, later announcing $n_t$26. In the asymptotic IID limit, the lower bounds on the secret-key rates are

$n_t$27

$n_t$28

In the ideal noiseless GMAC one finds $n_t$29 bits per channel use, and by time-sharing any point in the convex hull $n_t$30. The security proof uses purification of the channels, one-shot Renes-Renner lower bounds, Berta’s entropic uncertainty relation, and the asymptotic equipartition property (Das et al., 2021).

These two lines of work show that MCSC can mean either a network of adversarially corrupted classical links or a multipartite quantum communication process. In both cases, the number and type of channels directly determine which parties or senders retain provable security guarantees.

6. Hardware, application-layer, and heterogeneous-channel instantiations

Several recent works implement MCSC as a concrete system architecture. In a frequency-multiplexed THz OOK design, a reprogrammable amplitude-coding metasurface uses two layers of graphene with separate biasing voltages and supports two frequencies, $n_t$31 and $n_t$32. Four digital states $n_t$33, $n_t$34, $n_t$35, and $n_t$36 correspond to different $n_t$37 pairs and switch the effective reflection coefficients $n_t$38 and $n_t$39 between low and high states. The OOK on-off ratio is

$n_t$40

and in practice $n_t$41. The composite radiated field is

$n_t$42

and demultiplexing uses band-pass filters centered at $n_t$43 and $n_t$44. The scheme combines this with a substitution cipher and a block-to-channel assignment procedure; the summary states that without knowledge of the substitution key and block-to-channel mapping the attacker faces $n_t$45 possible plaintext mappings per block. Theoretical BER curves predict that at an average SNR of $n_t$46 dB and weak turbulence $n_t$47, BER falls below $n_t$48 (Farzin et al., 2024).

A related THz design uses polarization rather than frequency as the multichannel resource. The metasurface employs two graphene nanoribbons controlled by independent bias voltages $n_t$49 and $n_t$50, and at $n_t$51 the reflected field is synthesized as

$n_t$52

This enables two distinct digital streams, one on the $n_t$53-channel and one on the $n_t$54-channel, and supports a Double Random Phase Encryption protocol in which plaintext $n_t$55 is multiplied by two random phase masks with a Fourier transform in between. The same summary gives

$n_t$56

for two independent channels with bandwidth $n_t$57 and SNR$n_t$58 dB, and reports simulated BER$n_t$59–$n_t$60 at SNR$n_t$61 dB, while polarization crosstalk more than $n_t$62 dB down drives the intercepted SNR below $n_t$63 dB and BER to $n_t$64 (Farzin et al., 2024).

At the application layer, multichannel steganography formalizes MCSC as a hybrid of Cover–Synthesis and Cover–Modification,

$n_t$65

with three insecure channels $n_t$66. The sender transmits $n_t$67 over $n_t$68, $n_t$69 over $n_t$70, and $n_t$71 over $n_t$72. Security is analyzed through a PPT multichannel adversary game, and Theorem 1 states

$n_t$73

under the stated assumptions. The same reference reports average transmission and processing figures such as $n_t$74 s for Setup & Synth, $n_t$75 s and $n_t$76 s for Transmit Stego, and image-stego quality metrics PSNR $n_t$77 dB, MSE$n_t$78, SSIM$n_t$79 (Omego et al., 8 Jan 2025).

In IoT and WoT, MCSC is implemented as AES-protected dynamic channel hopping. The architecture includes an AES Encryption Module using AES-128, a channel manager over $n_t$80, a synchronization unit, a packetization module, and an nRF24L01+ wireless front-end. The hopping sequence is

$n_t$81

and synchronization adjusts the local clock when $n_t$82. The comparative table gives an error rate of $n_t$83 for MCSC, with jamming resilience $n_t$84, MITM resilience $n_t$85, and replay resilience $n_t$86, while synchronization overhead is reported as $n_t$87 instead of the $n_t$88–$n_t$89 of traditional multi-channel schemes. Under lab-simulated interference, the same summary reports $n_t$90 PDR and $n_t$91 ms average latency in low interference, $n_t$92 PDR and $n_t$93 ms average latency in high interference, and $n_t$94–$n_t$95 mAh per $n_t$96 hr under typical duty cycles (Barman et al., 11 Sep 2025).

Vehicular MCSC uses heterogeneous channels in series rather than parallel. In the proposed V2I authentication scheme, credentials are first exchanged over an IEEE 1609.2–compliant TLS link with

$n_t$97

and are then validated through an LOS optical challenge-response. The RSU sends a uniformly random $n_t$98-bit challenge $n_t$99 through the TLS channel; the vehicle maps it to a $n_e$00-bit frame $n_e$01 and re-emits it with its headlights using on-off keying over $n_e$02 flashes with $n_e$03 ms. A SlowFast two-stream convolutional architecture with a dual-channel design decodes the video response. The real-world results report best test accuracy $n_e$04 on the RC-car platform and $n_e$05 on the real-car platform, with average accuracy $n_e$06 and $n_e$07, respectively, and inference latency $n_e$08 ms and $n_e$09 ms (Vincenzi et al., 1 May 2025).

These applied systems show that MCSC is no longer confined to secrecy-capacity analysis. It has become a design pattern for combining multiple communication resources—frequency, polarization, logical cover channels, hopping channels, or LOS+NLOS modalities—to obtain operational security properties.

7. Recurring trade-offs, limits, and misconceptions

A recurrent misconception is that adding channels automatically yields secrecy. The antenna-scaling laws show otherwise. In MISOME, if $n_e$10, the eavesdropper can drive the secrecy capacity to zero even as $n_e$11 (0708.4219). In MIMOME, the zero-capacity condition is $n_e$12, and in the large-system limit the eavesdropper needs $n_e$13 times as many antennas as the sender and intended receiver have jointly to force zero secrecy (Khisti et al., 2010). Multi-channel structure therefore creates secure degrees of freedom only when the channel geometry is favorable.

A second recurring issue is the value of eavesdropper CSI. The MISOME high-SNR result shows that the penalty for not knowing the eavesdropper’s channel is small for masked beamforming, up to an SNR penalty of $n_e$14 (0708.4219). By contrast, the general MIMOME analysis states that a semi-blind “masked” MIMO transmission strategy can be arbitrarily far from capacity (Khisti et al., 2010). The distinction is structural: the rank-one MISOME geometry is much more forgiving than the full MIMOME geometry.

A third trade-off concerns robustness versus legitimate performance. In robust multicast MCC, adding AN slightly worsens legitimate BER but dramatically raises eavesdropper BER/MSE, and choosing $n_e$15 trades security versus transmit power while AN variance $n_e$16 trades legitimate BER versus eavesdropper MSE (Jagyasi et al., 2020). In MAWC-CM, private-sum rates lie below those of the compound MAC due to secrecy sacrifice, although when the eavesdropper’s channel is much noisier the secrecy-constraint region can exceed the compound-MAC region (Zivari-Fard et al., 2014).

Finally, protocol-level multichannel systems introduce their own overheads. Multichannel steganography lists increased complexity in synchronization and nonce management and higher overall latency due to three-phase exchange as explicit limitations (Omego et al., 8 Jan 2025). The IoT framework is organized around the opposite objective—reducing synchronization overhead to $n_e$17—which indicates that in practical MCSC, synchronization cost is itself a first-order security-performance variable (Barman et al., 11 Sep 2025).

Across these literatures, MCSC is best understood not as a single protocol class but as a unifying design principle: security is strengthened by distributing information, coding, or authentication evidence across multiple channels, while the actual gain depends on geometry, adversarial knowledge, synchronization cost, and the precise security criterion being enforced.