Normalized Channels in Fading MIMO

• Normalized Channels are defined as the asymptotic scaling laws of fading channel capacity, emphasizing the impact of temporal correlation, power constraints, and the absence of CSI.
• The framework establishes tight upper and asymptotic lower bounds under both sum and per-antenna power constraints, highlighting distinct signaling strategies for low-SNR regimes.
• Results extend to SISO channels with delay spread, illustrating structural parallels between spatial MIMO and temporal frequency-selective fading in terms of normalized capacity.

Low-SNR capacity of fading channels concerns the behavior of channel capacity when the signal-to-noise ratio tends to zero while the channel remains random. In "Low SNR Capacity of Fading Channels -- MIMO and Delay Spread," the setting is a discrete-time Rayleigh fading multiple-input multiple-output (MIMO) channel with no channel state information at the transmitter and receiver. The fading is assumed to be correlated in time and independent from antenna to antenna. Peak and average transmit power constraints are imposed either on the sum over antennas or on each individual antenna. Within this model, the work presents an upper bound and an asymptotic lower bound on channel capacity as the signal-to-noise ratio approaches zero, identifies the limit of normalized capacity under the sum power constraints and, for a subclass of channels, under individual power constraints, and states that these results carry over to a SISO channel with delay spread, i.e. frequency selective fading [0701078].

1. Problem statement and scope

The topic is the low-SNR asymptotics of fading-channel capacity under nontrivial input constraints and without channel state information at either end of the link. The emphasis is not on a static deterministic channel but on a stochastic fading law, and not on high-SNR multiplexing behavior but on the regime in which the signal-to-noise ratio approaches zero [0701078].

The principal elements of the formulation can be summarized as follows.

Aspect	Specification
Channel class	Discrete-time Rayleigh fading MIMO
CSI assumption	No channel state information at the transmitter and receiver
Fading law	Correlated in time and independent from antenna to antenna
Input constraints	Peak and average transmit power constraints
Constraint variants	Sum over antennas, or each individual antenna
Main asymptotic regime	Signal-to-noise ratio approaches zero
Extension	SISO channel with delay spread, i.e. frequency selective fading

This scope places the work at the intersection of fading-channel information theory, low-power communication, and asymptotic capacity analysis. The combination of temporal correlation, spatial multiplicity, and the absence of channel state information makes the problem structurally different from coherent MIMO capacity analysis.

2. Channel model and fading assumptions

The channel model is discrete-time, Rayleigh fading, and multiple-input multiple-output. The fading process is assumed to be correlated in time, so the channel law is not memoryless across channel uses. At the same time, the fading is independent from antenna to antenna, so the spatial dependence structure is simpler than the temporal one [0701078].

These assumptions separate two sources of complexity. Temporal correlation means that the capacity analysis must account for fading memory. Antenna-wise independence means that the spatial branches do not share fading randomness directly, even though they are jointly constrained through the power budget. This combination is especially relevant in the low-SNR regime, where the fine structure of channel statistics can dominate first-order scaling.

A further defining assumption is the absence of channel state information at both transmitter and receiver. Under that restriction, capacity is governed by the statistical description of the fading process and by the admissible signaling strategies under peak and average power constraints, rather than by instantaneous adaptation to realized channel states.

3. Power constraints and the role of normalization

The paper imposes both peak and average transmit power constraints, and studies two distinct modes of enforcement: constraints on the sum over antennas, and constraints on each individual antenna [0701078]. This distinction is central, because it separates a pooled-power architecture from a per-antenna-limited architecture.

Under sum power constraints, the power limitation is applied collectively across the transmit array. Under individual power constraints, each antenna is separately constrained. This suggests two different optimization geometries for low-SNR signaling. In the first, spatial degrees of freedom can be coordinated under a common budget; in the second, each branch is locally limited, which can alter both the achievable signaling structure and the asymptotic scaling law.

The work also identifies a limit of normalized capacity under the sum power constraints and, for a subclass of channels, under individual power constraints [0701078]. The published description refers to “normalized capacity” but does not specify the exact normalization. A plausible implication is that the analysis isolates a scaling law for capacity in the vanishing-SNR regime rather than only giving finite-SNR inequalities.

4. Low-SNR bounds and asymptotic capacity characterization

The main analytical results are an upper bound and an asymptotic lower bound on channel capacity as the signal-to-noise ratio approaches zero [0701078]. In an asymptotic information-theoretic analysis, that pairing is significant: the upper bound limits what any admissible signaling scheme can achieve, while the asymptotic lower bound demonstrates attainability to leading order by an explicit or implicit family of inputs.

The abstract further states that the limit of normalized capacity is identified under the sum power constraints, and, for a subclass of channels, for individual power constraints [0701078]. This is stronger than merely showing matching rates up to order symbols; it indicates that a definite asymptotic limit is obtained in those cases. The restriction to “a subclass of channels” under individual power constraints indicates that the per-antenna-constrained case is structurally more delicate.

Taken together, these statements indicate a hierarchy of results. The broadest results are low-SNR upper and asymptotic lower bounds for both power-constraint models. A sharper asymptotic identification is available under sum power constraints. A similarly sharp identification under individual power constraints is available only for a subclass of channels. This suggests that spatial power pooling materially simplifies the asymptotic characterization.

5. Extension to delay-spread SISO channels

The paper states that its results carry over to a SISO channel with delay spread, i.e. frequency selective fading [0701078]. This extension is important because it links a spatially distributed model, MIMO fading, to a temporally dispersive single-antenna model.

A SISO channel with delay spread introduces frequency selectivity through multipath structure rather than through multiple transmit and receive antennas. The carry-over result therefore indicates that the low-SNR asymptotic arguments are not confined to spatial multiplexing architectures. Instead, they apply as well to a channel whose essential complication is temporal dispersion.

This suggests a structural parallel between spatial diversity under antenna indexing and temporal diversity under delay indexing. In both cases, the low-SNR analysis appears to depend on how the stochastic channel law interacts with peak and average power constraints when channel state information is unavailable.

6. Significance within asymptotic capacity research

The work belongs to a broader family of asymptotic channel analyses that replace exact finite-parameter expressions with limits, bounds, or deterministic approximations. In a different asymptotic direction, later work on correlated-fading MIMO multiple-access channels derived deterministic approximations of normalized mutual information, normalized sum-rate with MMSE detection, and MMSE SINR as all system parameters grow large at the same speed (Hoydis et al., 2011). That large-system regime is distinct from the vanishing-SNR regime of the present topic, but both use normalization to extract stable asymptotic structure from otherwise difficult channel models.

Another related asymptotic theme appears in Gaussian channels, where a differential relation between capacity and a quantity termed normalized optimal detection error was established for vector Gaussian channels and extended asymptotically to continuous-time Gaussian channels (Hammerich, 2017). That line of work is not a fading-channel low-SNR result, but it illustrates how normalization can expose first-order behavior of information measures under limiting operations.

Within that broader landscape, the distinctive contribution of the 2007 work is its focus on Rayleigh fading MIMO without channel state information at either transmitter or receiver, with temporal correlation, peak and average power constraints, and a direct extension to delay-spread SISO channels [0701078]. Its central subject is not large-dimensional randomness or detection-theoretic sensitivity, but the limiting capacity law of fading channels at vanishing signal-to-noise ratio.

7. Conceptual interpretation and limitations of the available statement

At the level of the published description, the key conceptual message is that low-SNR capacity in fading channels is strongly shaped by three ingredients taken jointly: channel memory through temporal correlation, the geometry of the transmit-power constraint, and the absence of channel state information [0701078]. The distinction between sum and individual power constraints is not incidental; it is built directly into the asymptotic capacity characterization.

The same description, however, leaves several technical components unspecified. It does not provide the explicit form of the upper bound, the asymptotic lower bound, the normalized-capacity limit, or the precise subclass of channels for which the individual-constraint limit is identified. It also does not state the exact signaling constructions or converse techniques. Accordingly, the topic is best understood at this level as an asymptotic information-theoretic framework with clearly defined assumptions and result classes, rather than as a fully formulaic capacity theorem in the absence of the complete paper text.

Even with that limitation, the topic remains well defined. It concerns discrete-time Rayleigh fading MIMO channels, no channel state information at the transmitter and receiver, fading correlated in time and independent from antenna to antenna, peak and average transmit power constraints under both sum and individual formulations, upper and asymptotic lower capacity bounds as the signal-to-noise ratio approaches zero, identification of normalized-capacity limits in specified cases, and transfer of the results to SISO delay-spread channels [0701078].