Quasi-Static Channel Model

• QSCM is defined by channel gains drawn at the start of a block that remain constant, shifting focus from ergodic averaging to outage performance.
• It applies to diverse scenarios — from SIMO fading and MIMO-OFDM to covert and resonant beam communications — while handling frequency selectivity and spatial structures.
• Analytical techniques such as meta-converse bounds, MMSE estimation, and finite-blocklength coding illustrate vanishing dispersion and substantial blocklength reductions.

Searching arXiv for recent and foundational papers on quasi-static channel models and related finite-blocklength/fading formulations. The quasi-static channel model (QSCM) is a block-fading formulation in which the channel gains or channel matrices are drawn once at the beginning of a codeword, slot, or packet and remain fixed throughout the entire transmission interval; equivalently, the coherence time satisfies $T_c \ge n$. Across the cited literature, this assumption appears in single-input multiple-output fading, multi-antenna covert communication, MIMO-OFDM channel estimation, massive MIMO unsourced random access, and resonant beam communication. Its defining operational consequence is the absence of time diversity within the block, so each transmission experiences one random channel realization over the whole coding interval (Yang et al., 2013, Liu et al., 31 Mar 2026, Hoang et al., 11 Jul 2025, Gkagkos et al., 2022, Li et al., 2024).

1. Defining property and temporal structure

In the most direct formulation, quasi-static means that the fading coefficient or channel matrix is sampled once per block and then frozen for all channel uses in that block. For SIMO fading, the vector $\mathbf h$ is drawn once per block of length $n$ and remains constant over the block; for quasi-static MIMO covert communication, the matrices $H_b$ and $H_w$ are drawn once and remain exactly the same over the entire codeword of length $n$; for quasi-static massive MIMO unsourced random access, each user channel vector $h_k$ stays fixed over all $n$ channel uses of the packet; and for the OFDM source-domain QSCM, the channel is constant over one resource block of $M$ OFDM symbols and changes independently from one slot to the next (Yang et al., 2013, Liu et al., 31 Mar 2026, Gkagkos et al., 2022, Hoang et al., 11 Jul 2025).

This temporal invariance does not force a single spatial or spectral structure. In the OFDM setting, the model is explicitly frequency-selective, because the frequency response is built from a sum of discrete multipath components with delays, phases, and array responses, while the Doppler shifts are set to zero within the slot. In the resonant beam setting, the quasi-static scenario is defined physically by relatively fixed transmitter and receiver locations and no beam wandering; the resulting channel is initially described as a discrete-time Markov channel across reflection rounds rather than a memoryless fading law (Hoang et al., 11 Jul 2025, Li et al., 2024).

A common misunderstanding is to equate quasi-static with flat fading or with a memoryless channel law. The cited formulations show otherwise: quasi-static refers to temporal constancy over the block, not to the absence of frequency selectivity, not to scalarity, and not necessarily to memorylessness.

2. Canonical mathematical formulations

The literature uses several mathematically distinct QSCMs, all sharing blockwise channel constancy.

Setting	Input-output relation	Frozen state over the block
SIMO fading	$$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$	$\mathbf h$0
Covert MIMO fading	$\mathbf h$1	$\mathbf h$2
MIMO-OFDM QSCM	$\mathbf h$3	$\mathbf h$4 identical across $\mathbf h$5 in one slot
Massive MIMO URA	$\mathbf h$6	user channels $\mathbf h$7
Resonant beam quasi-static scenario	$\mathbf h$8 for $\mathbf h$9	round-to-round optical state under fixed geometry

In the SIMO fading formulation, the received matrix satisfies

$n$0

with a per-codeword peak power constraint $n$1. In massive MIMO unsourced random access, the received signal is

$n$2

with $n$3 fixed across the packet. In covert multi-antenna fading, Alice sends an $n$4 matrix $n$5, Bob and Willie observe the corresponding matrix channels, and Willie is assumed to know $n$6 perfectly in all CSI cases (Yang et al., 2013, Gkagkos et al., 2022, Liu et al., 31 Mar 2026).

The OFDM source-domain QSCM makes the frequency-selective structure explicit: $n$7 and because $n$8, all $n$9 in the same slot share identical $H_b$0. This formulation shows that QSCM can be frequency-selective and array-aware while still being quasi-static in time (Hoang et al., 11 Jul 2025).

The fading laws used under QSCM are likewise heterogeneous. The cited works include Rayleigh, Rician with $H_b$1-factor, and Nakagami-$H_b$2 examples, and in the covert MIMO setting the only worst-case assumption needed for the warden channel is $H_b$3 for a known $H_b$4 (Yang et al., 2013, Liu et al., 31 Mar 2026).

3. Outage capacity, finite blocklength, and vanishing dispersion

For quasi-static SIMO fading, the central performance object is outage rather than ergodic averaging. Writing $H_b$5, the outage event at rate $H_b$6 is

$H_b$7

and the outage capacity, or $H_b$8-capacity, is

$H_b$9

Under mild smoothness conditions on the pdf of $H_w$0 and assuming $H_w$1 is a growth point of the outage-cdf, the quasi-static SIMO model has zero $H_w$2-dispersion: $H_w$3 This contrasts with the usual normal approximation for ergodic channels, where a nonzero $H_w$4 term governed by channel dispersion appears (Yang et al., 2013).

The derivation combines a converse and an achievability result that match up to $H_w$5. The converse applies the meta-converse theorem with auxiliary channel

$H_w$6

and an asymptotic Cramér-Esseen analysis yields $H_w$7. The achievability part uses a $H_w$8–$H_w$9 bound on a degraded test, codewords on the $n$0-sphere $n$1, and a geometric angle-test decoder, giving

$n$2

Together these establish that the usual $n$3 backoff disappears in the quasi-static regime (Yang et al., 2013).

The numerical illustration for a particular $n$4 SIMO Rician channel makes the phenomenon concrete. With two independent Rician channels each with $n$5 dB and target error $n$6, the finite-blocklength achievability and converse bounds show that achieving $n$7 of $n$8 requires about $n$9 when CSIT+CSIR are available and $h_k$0 with no CSI at transmitter or receiver, whereas an AWGN channel with the same capacity would require $h_k$1. This suggests that, in some quasi-static operating points, the vanishing-dispersion effect translates into an order-of-magnitude blocklength reduction (Yang et al., 2013).

4. CSI, feedback, and coding architectures

One major line of work studies how QSCM interacts with nonasymptotic coding when CSI is imperfect or feedback is structured. In the real quasi-static fading model

$h_k$2

with $h_k$3 fixed over the block, imperfect CSIT is represented by an estimate $h_k$4 and distortion $h_k$5 subject to $h_k$6. The associated feedback channel is quantized and additive in modulo-lattice form. To adapt the classical Schalkwijk-Kailath scheme, the encoder-decoder introduces a modulo-lattice function on the feedback path and an auxiliary subtraction in the forward decoder. On the no-aliasing event, the transmitter recovers a feedback quantity of the form $h_k$7, and the receiver cancels the known offset $h_k$8 to form

$h_k$9

after which MMSE updating proceeds. The paper states that the decoding error caused by the imperfect CSIT can be perfectly eliminated, and gives the achievable-rate bound

$n$0

with $n$1. Complexity remains extremely low, with linear updates and one scalar modulo per round (Yang et al., 2 Jul 2025).

A distinct CSI question arises in covert quasi-static multi-antenna fading. The model distinguishes four cases, $n$2, corresponding to no-CSI, CSIT only, CSIR only, and CSIT+CSIR, while Willie always knows $n$3 perfectly. Reliability is constrained by $n$4, covertness by the KL divergence condition

$n$5

and transmit power must decay as $n$6. The resulting finite-blocklength covert rate obeys

$n$7

with

$n$8

Unlike the non-covert quasi-static MIMO result of Yang et al. cited in the paper, CSI availability at Alice or Bob does not improve the first-order covert rate, because the covertness constraint forces the power to vanish too quickly to exploit water-filling (Liu et al., 31 Mar 2026).

5. Estimation, domain adaptation, and multiuser processing

In MIMO-OFDM channel estimation, QSCM serves as a source-domain simulation model. The assumptions are block-fading over one resource block of $n$9 OFDM symbols, zero Doppler within the slot, and a multipath representation with gains, delays, phases, and steering vectors. The resulting received signal on each subcarrier and OFDM symbol is

$M$0

Using the DeepMIMO O1 dataset at $M$1 GHz, with $M$2 channel realizations, $M$3 subcarriers, $M$4 symbols, and DM-RS pilot pattern “type 1/A,” the estimation pipeline begins with LS or LS-LI initialization and then applies either CNN or GAN refinement. The discrepancy between the QSCM source domain and the realistic MBCM target domain is quantified by a Wasserstein-1 distance approximated via Sinkhorn to $M$5. Without adaptation, CNN and GAN degrade NMSE relative to LS-LI under domain shift; after fine-tuning, LS-LI-CNN and LS-LI-GAN surpass LS-LI by $M$6 dB NMSE at low SNR, while at high SNR all methods converge (Hoang et al., 11 Jul 2025).

This source-domain use of QSCM is notable because it separates temporal quasi-staticity from environmental realism. QSCM sets $M$7 and draws idealized ray parameters from the DeepMIMO distribution, whereas the target MBCM uses OSM map plus RayTracing, CDL clusters, and nonzero Doppler. The paper states the boundary condition succinctly: at zero Doppler and in an idealized environment, $M$8 (Hoang et al., 11 Jul 2025).

In quasi-static massive MIMO unsourced random access, the same blockwise constancy enables a different receiver architecture. The pilot block satisfies

$M$9

and the MMSE channel estimator from pilots alone is

$$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$0

Because $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$1 is fixed over the block, tentative codeword decisions can be fed back into a second MMSE-like update,

$$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$2

which supports iterative estimation and decoding. Activity detection is likewise based on blockwise energy correlation through statistics $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$3, and the design criterion is to minimize $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$4 subject to $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$5. The paper emphasizes both sides of the tradeoff: quasi-staticity reduces pilot overhead and permits full-block averaging gain, but as the number of active users grows, pilot collisions and interference still raise the required $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$6 (Gkagkos et al., 2022).

6. Variant channel reductions, applications, and limitations

The quasi-static resonant beam communication model illustrates that QSCM can arise in non-RF settings and can coexist with internal channel memory. In the quasi-static scenario, transmitter and receiver remain essentially fixed, the optical resonator is pumped until a steady circulating beam of power $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$7 is reached, and communication then proceeds in reflection rounds. The beam amplitude process obeys

$$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$8

so that, for fixed $$\mathbf Y=\mathbf x\,\mathbf h^{\mathsf H}+\mathbf W$$9, $\mathbf h$00 is a discrete-time Markov chain. By choosing

$\mathbf h$01

to force the instantaneous channel gain to a constant $\mathbf h$02, the model is flattened into the memoryless amplitude-constrained AWGN channel

$\mathbf h$03

Capacity upper and lower bounds are then optimized by a bisection plus exhaustive-search procedure over $\mathbf h$04, $\mathbf h$05, and $\mathbf h$06, and the numerical results report that the optimized bounds are virtually indistinguishable over all $\mathbf h$07 (Li et al., 2024).

Several limitations recur across QSCM formulations. The most explicit is the loss of time-domain diversity: in quasi-static links, every codeword sees a single random SNR gain, and in massive MIMO unsourced random access a deep fade lasting the entire block cannot be averaged out. If the channel changes within the packet or slot, more frequent retraining is required and the blockwise MMSE updates cease to be valid (Yang et al., 2013, Gkagkos et al., 2022). In covert communication, a further limitation is that power must scale as $\mathbf h$08, so the first-order rate follows the square-root law rather than a nonvanishing outage capacity (Liu et al., 31 Mar 2026).

A second recurring misconception is that more CSI necessarily improves the leading finite-blocklength behavior. In non-covert quasi-static SIMO, zero dispersion holds regardless of whether fading realizations are available at the transmitter and/or the receiver under mild conditions on the channel gains. In covert quasi-static MIMO, CSI at Alice or Bob does not affect the finite blocklength performance. Conversely, in the feedback setting with imperfect CSIT, careful architectural modification is needed before feedback can recover Schalkwijk-Kailath-type reliability (Yang et al., 2013, Liu et al., 31 Mar 2026, Yang et al., 2 Jul 2025).

Taken together, these results indicate that QSCM is best understood not as a single channel law but as a modeling regime defined by blockwise channel constancy. Within that regime, the main analytical objects shift from ergodic averaging to outage, from time diversity to spatial structure, and from channel tracking to one-shot inference or blockwise adaptation. A plausible implication is that the practical value of QSCM depends less on any single fading distribution than on whether the entire signaling interval genuinely fits inside one coherence interval and whether the application can tolerate the resulting outage-dominated behavior.