🔢 mathematics

Probabilistic Temporal Shaping for Level-Constrained Signaling on Bandlimited Additive White Gaussian Noise Channels

This paper derives new lower bounds on the capacity of bandlimited AWGN channels with level-constrained inputs by introducing probabilistic temporal shaping, which improves existing high-SNR performance by at least 1.94 dB while ensuring signal invertibility.

Original authors: Mahdi Mahvari (Shitz), Gerhard Kramer (Shitz), Shlomo Shamai (Shitz)

Published 2026-08-11

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Mahdi Mahvari (Shitz), Gerhard Kramer (Shitz), Shlomo Shamai (Shitz)

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to send a secret message across a noisy room using only a flashlight. In the real world, the devices that turn our digital "ones and zeros" into these physical signals—like the flashlight beam—are expensive and hungry for power. The more precise the signal needs to be (the more "colors" of light you can switch between), the more energy it takes to create it. To save energy, engineers often try to simplify the signal, forcing it to be very simple, like a flashlight that can only be fully ON or fully OFF, never in between. This is called "level-constrained" signaling.

However, there's a catch. When you send these simple ON/OFF signals through a medium like a wire or the air, the signal gets smeared out, like a drop of ink spreading in water. This smearing makes it hard for the receiver to tell exactly when the signal switched from OFF to ON. If the signal is too messy, the message gets lost in the noise. Scientists have been trying to figure out the absolute maximum amount of information they can squeeze through this "smeared" channel without using too much power. It's a bit like asking: "What is the fastest speed a car can drive on a muddy road without getting stuck, if it can only use two gears?"

This paper tackles that exact puzzle. The researchers, Mahdi Mahvari, Gerhard Kramer, and Shlomo Shamai, looked at a specific type of communication channel where the signal is limited to just two levels (like a flashlight that is either bright or dark) and the path it travels is "bandlimited," meaning it can't change too quickly. They wanted to find a smarter way to arrange the timing of these ON/OFF switches to send more data.

The team discovered that the best way to do this isn't to change the intensity of the light (which they proved doesn't help in this specific setup), but to change the timing of the switches in a very clever, probabilistic way. They developed a method called "Probabilistic Temporal Shaping" (PTS). Think of it like a drummer who doesn't just keep a steady beat, but varies the time between drum hits based on a complex, optimized pattern. By carefully choosing when to switch the signal on and off, rather than just how often, they found a way to pack more information into the same space.

Their calculations show that this new method is significantly better than the previous best-known methods. At high signal quality, their new approach improves the data rate by at least 1.94 decibels (dB) compared to the old record holder. They also found a slightly simpler version of their method that still gains a solid 1.64 dB. While they didn't prove this is the absolute theoretical limit of what's possible, they did prove that their new method is mathematically superior to the schemes used before, effectively pushing the boundaries of how much data we can send using simple, power-efficient signals.

The Story of the Flashlight and the Muddy Road

Let's dive deeper into how they cracked the code.

The Problem: The "Smear" and the "Switch"
Imagine you are sending a message by flipping a light switch. In a perfect world, the light would go from OFF to ON instantly. But in the real world, the "muddy road" (the bandlimited channel) smears that instant change. The light takes a moment to brighten up, and the receiver might get confused about exactly when the switch was flipped.

To make things harder, the paper focuses on a strict rule: the signal can only be at one of two levels (ON or OFF). You can't dim the light to 50% to send more information. You have to work with just two states. The researchers looked at signals that switch states only once per "Nyquist-rate sample" (a fancy way of saying "once per the fastest possible time slot allowed by the channel").

What They Ruled Out: Don't Wiggle the Volume
First, the authors tested an idea that might seem obvious: "What if we vary the brightness?" Maybe sometimes the light is bright, sometimes dim, to encode extra data?
They did the math and found a surprising result: No. In this specific setup, changing the amplitude (brightness) of the signal does not help increase the amount of information you can send. The paper explicitly shows that sticking to the strict "ON or OFF" levels is actually the best strategy for maximizing the information capacity in this scenario. If you try to wiggle the volume, you just waste potential.

The Solution: The Dance of the Switches (PTS)
So, if we can't change the brightness, how do we send more data? We change the timing.

The researchers introduced a technique called Probabilistic Temporal Shaping (PTS).
Imagine a drummer again. The old way of sending data was like a drummer who hits the drum at perfectly regular intervals, or perhaps with a simple, repetitive pattern. The new method is like a drummer who follows a complex, optimized rhythm. The drummer doesn't just hit the drum; they decide exactly when to hit it based on a specific probability map.

In their model, the "switching times" (when the signal flips from ON to OFF) are chosen from a specific distribution. Instead of picking times randomly or in a fixed pattern, they calculated the perfect distribution of times that would make the signal easiest to decode at the other end.

They found that the optimal way to pick these times is related to a mathematical shape involving sine waves and determinants (a type of matrix calculation). It's a bit like finding the perfect spacing for stepping stones across a river so you don't slip. If you space them too evenly, the current (noise) knocks you off. If you space them randomly, you might step in the water. But if you space them according to their special "optimal density," you cross the river with the least amount of trouble.

The Results: A Clear Win
The team ran the numbers to see how much better this new "dance" was compared to the old ways.

The Old Champion: The best previous method (called "Scheme D" in the paper) had a certain efficiency limit.
The New Champion: The new PTS method, using the optimal timing density, pushed that limit higher.
The Gain: At high signal-to-noise ratios (when the connection is good), the new method improves the capacity by 1.94 dB.
The Simpler Version: They also tested a "sequential" version of the method, which is easier to build in real life. This simpler version still managed to gain 1.64 dB over the old best.

To put those numbers in perspective, in the world of wireless communication, a gain of even a fraction of a decibel is huge. A gain of nearly 2 dB is a massive leap forward.

How Sure Are They?
The authors didn't just guess; they proved it.

They mathematically proved that changing the amplitude (brightness) doesn't help.
They derived the exact mathematical formula for the best timing distribution.
They used simulations and mathematical theorems (like Szegő's theorem for matrices) to calculate the exact improvement.
They showed that as the number of signal switches increases, the improvement gets closer and closer to a theoretical limit of about 0.4053 (represented as a power factor $\gamma_1$ ), which is significantly higher than the previous best of 0.2586.

The Takeaway
This paper doesn't claim to have solved the entire mystery of communication forever. In fact, the authors note that their method assumes the signal switches only once per time slot. They suspect that if you allow signals to switch more often, you might find even better ways to send data. But for the specific problem of "level-constrained" signals (ON/OFF only) with one switch per slot, they have found a much smarter way to dance the rhythm. By letting the timing of the switches follow a carefully calculated, probabilistic pattern, we can squeeze significantly more information through the same noisy, power-hungry channels. It's a reminder that sometimes, the secret to speed isn't working harder, but timing your moves just right.

Technical Summary: Probabilistic Temporal Shaping for Level-Constrained Signaling on Bandlimited AWGN Channels

Problem Formulation
This paper investigates the capacity of bandlimited additive white Gaussian noise (AWGN) channels subject to level constraints, where the input signal is restricted to two discrete amplitude levels, $x(t) = \pm \sqrt{P}$ . This model is relevant to scenarios involving one-bit quantized signaling, such as in digital-to-analog converters (DACs) and power amplifiers (PAs) where high resolution is power-prohibitive or where peak-to-average power ratio (PAPR) must be minimized. The channel is modeled as an ideal low-pass filter with bandwidth $W$ followed by AWGN. While the capacity of unconstrained Gaussian channels is well-known, the capacity under level constraints ( $C_L$ ) for bandlimited channels remains difficult to compute. Existing literature provides bounds of the form $C_L/W \approx \log(1 + \gamma \text{SNR})$ , where the power factor $\gamma$ is significantly less than 1. The primary objective is to derive tighter lower bounds on this capacity.

Methodology
The authors focus on signaling schemes where the input signal $x(t)$ exhibits at most one data-dependent sign change per Nyquist-rate sampling interval ( $T = 1/2W$ ). This constraint ensures the mapping from the sign-change times to the channel output is invertible.

Amplitude Modulation Analysis: The paper first addresses whether introducing amplitude modulation (varying the magnitude of the steps between $\pm \sqrt{P}$ ) can improve capacity. By analyzing the differential entropy of the noise-free channel output, $h(Z)$ , the authors prove that for schemes with at most one sign change per interval, amplitude modulation does not increase entropy. Consequently, level-constrained signaling (fixed amplitude) maximizes $h(Z)$ under both peak and average power constraints.
Probabilistic Temporal Shaping (PTS): The core methodology involves optimizing the probability density function (PDF) of the sign-change times, denoted as $c = (c_1, c_2, \dots, c_N)$ .
- The authors derive the optimal shaping density $p^*(c)$ that maximizes the mutual information. This density is shown to be proportional to the absolute value of the determinant of the Jacobian matrix of the channel mapping, $| \det S_c |$ .
- Using properties of Cauchy matrices and Vandermonde determinants, the authors derive a closed-form expression for the determinant involving sine functions.
- A theorem is presented (Theorem 2) that expresses the maximum entropy $h(Z)$ in terms of the determinant of a skew-symmetric matrix $\Omega$ constructed from modified Sine integrals. This allows for the calculation of the power factor $\gamma_1$ for arbitrary block lengths $N$ .
Sequential PTS Scheme: Recognizing that computing the optimal joint density for large $N$ is computationally intensive, the authors propose a simpler "sequential" PTS scheme. This approach optimizes a unit-memory conditional density $p(c_i | c_{i-1})$ using coordinate descent and Monte Carlo integration. The support of this density is restricted to the interval $[c_{i-1}, i+1)$ , and the density is approximated by linear interpolation over sub-intervals.

Key Contributions and Results

Optimality of Level Constraints: The paper rigorously demonstrates that for the specific class of signals with one sign change per Nyquist interval, varying the amplitude (amplitude modulation) offers no capacity gain over fixed-level signaling.
Improved Lower Bounds: By applying the derived optimal PTS density, the authors calculate a new power factor $\gamma_1$ $γ_{1}$ .
- For a block length of $N=1000$ , the calculated $\gamma_1 \approx 0.4039$ .
- Asymptotic analysis using Szegő's theorem suggests $\lim_{N \to \infty} \gamma_1 = 4/\pi^2 \approx 0.4053$ .
Performance Gains:
- The optimal PTS scheme improves the best existing lower bound (Scheme D from prior work [10], where $\gamma_1 = 0.2586$ ) by at least 1.94 dB at high signal-to-noise ratio (SNR).
- The simpler sequential PTS scheme achieves a power factor of $\gamma_1 \approx 0.3773$ , yielding a gain of 1.64 dB over the previous best bound.
Comparison: The results are visualized in spectral efficiency vs. SNR plots, showing the new bounds significantly outperforming previous schemes (A, B, C, and D from [10]) and approaching the theoretical upper bound ( $\gamma_2 = 0.9259$ ) more closely than before.

Significance
The paper claims that Probabilistic Temporal Shaping is a powerful technique for enhancing the capacity of level-constrained bandlimited channels. By optimizing the timing of sign changes rather than just their occurrence, the system can better exploit the channel's degrees of freedom. The authors note that while their results represent a significant improvement over existing lower bounds, they are based on signals with an average of one sign change per Nyquist sample. They modestly suggest that schemes allowing for a larger number of sign changes per interval may yield further improvements, but their current work establishes a new benchmark for the single-sign-change regime. The findings are particularly relevant for the design of energy-efficient communication systems utilizing low-resolution DACs and PAs.

The Story of the Flashlight and the Muddy Road

More like this