Target-Rate Least-Squares Power Allocation over Parallel Channels

This paper proposes an efficient algorithm for minimizing total squared rate deviation in parallel Gaussian channels under a sum-power constraint by deriving a closed-form solution using the Lambert W function and a one-dimensional bisection method, which outperforms classical waterfilling and general-purpose solvers in both speed and target-tracking accuracy.

The Gist

Imagine you are the conductor of a massive orchestra, but instead of violins and trumpets, you have 8 to 1,000 different radio channels (like the tiny slices of frequency used by your Wi-Fi or 5G phone). Your goal is to hand out a limited amount of "energy" (power) to each instrument so they all play at the perfect volume.

In the old days, the conductor's rule was simple: "Give the loudest instruments as much power as possible." This is called Waterfilling. If you have a bucket of water (power) and a landscape of hills (channels), you pour the water in. It naturally fills the deep valleys (weak channels) first, but if you have enough water, it spills over the tops of the highest peaks (strong channels), making them blast even louder. The goal was just to make the total sound as loud as possible.

But here's the problem: In the real world, we don't always want the loudest possible sound.

• Maybe a video call needs exactly 3 megabits per second to look good.
• Maybe a smart sensor only needs 1 megabit to send a temperature reading.
• If you blast a channel way past its need, you aren't just wasting energy; you might be causing interference or wasting battery life.

This paper introduces a new, smarter way to conduct the orchestra called Target-Rate Least-Squares Power Allocation.

The New Philosophy: "Just Enough, Not Too Much"

Instead of trying to make the whole orchestra as loud as possible, this new method asks: "How close can we get every single instrument to its specific target volume?"

Think of it like a tailor fitting suits for a group of people:

• Person A needs a size 40 suit.
• Person B needs a size 32 suit.
• Person C needs a size 38 suit.

The Old Way (Waterfilling): The tailor gives everyone a giant size 50 suit because it's the most "efficient" way to use the fabric. Everyone is drowning in fabric, and the small people are tripping over it. It's wasteful and uncomfortable.

The New Way (This Paper): The tailor measures everyone and cuts the fabric to fit exactly their size.

• If you have enough fabric for everyone, you cut perfect suits for all.
• Crucially: If you run out of fabric, you don't just give the extra scraps to the biggest guy. You distribute the remaining fabric to get everyone as close as possible to their target size.
• The "No-Overshoot" Rule: The tailor will never make a suit bigger than the person needs. If a person needs a size 32, you never give them a 34. Any extra fabric is saved for the people who are still too small.

The Magic Trick: The "Lambert W" Function

You might think, "Okay, but how do you calculate exactly how much fabric to give 1,000 people at once without spending hours?"

Usually, solving this kind of math problem requires a supercomputer to guess and check millions of times. It's slow.

The authors of this paper found a mathematical shortcut. They discovered that the answer involves a special, somewhat mysterious function called the Lambert W function.

Think of the Lambert W function as a magic calculator button.

• In the old days, to figure out the power, you had to climb a mountain of math step-by-step.
• With this new method, you just press the "Lambert W" button, and it instantly tells you the exact power needed for each channel based on a single "tuning knob" (a variable called $\lambda$).

Why This Matters (The Results)

The paper tested this new method against the old "Waterfilling" method and found some amazing things:

1. It's a Speed Demon: When they tested it on 1,024 channels (like a modern 5G system), their new method was 1,890 times faster than the standard computer solvers. It's the difference between waiting for a slow dial-up internet connection and having 5G.
2. It Saves Energy: Because it follows the "No-Overshoot" rule, if the targets are easy to meet, it simply turns off the extra power. It doesn't waste energy blasting a channel that is already happy.
3. It's Fairer to the Weak: In the old method, strong channels got all the power. In this new method, the weak channels get a fair share of the power to help them reach their targets, ensuring no one is left behind.

The Bottom Line

This paper gives us a new tool for managing wireless networks. Instead of blindly pouring power into the strongest signals, we can now intelligently distribute power to hit specific goals.

• Old Way: "Make everything as loud as possible, even if it's wasteful."
• New Way: "Hit the target volume for everyone, save energy if we can, and do it so fast we can do it in real-time."

It's like switching from a sledgehammer to a precision scalpel. And thanks to the "magic button" (Lambert W function), we can use that scalpel instantly, even on massive networks.

Technical Summary

Here is a detailed technical summary of the paper "Target-Rate Least-Squares Power Allocation over Parallel Channels" by Bhaskar Krishnamachari.

1. Problem Formulation

The paper addresses the problem of power allocation in a system of $N$ parallel Gaussian channels (e.g., OFDM subcarriers), where the objective is not to maximize total throughput, but to track specific target spectral efficiencies ($T_i$) for each channel.

• System Model:
  • $N$ parallel channels with known channel gain-to-noise ratios $a_i > 0$.
  • Achievable rate on channel $i$: $r_i(P_i) = \log_2(1 + a_i P_i)$.
  • Each channel has a target rate $T_i \geq 0$.
  • Total power budget constraint: $\sum P_i \leq P_{tot}$.
• Objective:
  Minimize the sum of squared deviations between achieved rates and targets:
  $$\min_{\{P_i\}} J(\mathbf{P}) = \sum_{i=1}^N \left( \log_2(1 + a_i P_i) - T_i \right)^2$$
  subject to $\sum P_i \leq P_{tot}$ and $P_i \geq 0$.

This formulation differs from classical Waterfilling (which maximizes sum rate and always exhausts the power budget) and Margin-Adaptive allocation (which treats targets as hard constraints). Instead, it uses a "soft" least-squares penalty, allowing for smooth degradation when the power budget is insufficient to meet all targets.

2. Key Methodological Insights

The authors derive a closed-form solution using Karush-Kuhn-Tucker (KKT) conditions and the Lambert W function. The derivation relies on several critical structural properties:

A. The "No-Overshoot" Property

Theorem 4.1: The optimal solution never allocates power such that a channel's rate exceeds its target ($r_i \leq T_i$).

• Implication: If a channel exceeds its target, power can be reduced to exactly meet the target, lowering the squared error to zero and saving power for other channels. This fundamentally distinguishes the solution from waterfilling, which often "overshoots" targets on strong channels.

B. Convexity on a Restricted Domain

While the squared error of a concave log function is not globally convex, the authors prove that the optimal solution lies within a restricted domain $0 \leq P_i \leq \bar{P}_i$, where$\bar{P}_i = (2^{T_i}-1)/a_i$ is the power required to exactly meet the target.

• Within this domain, the objective function is convex, ensuring that KKT conditions are sufficient for global optimality.

C. Two Operating Regimes

The solution exhibits two distinct behaviors based on the power budget:

1. Case A (Sufficient Power): If $P_{tot} \geq \sum \bar{P}_i$, the optimal strategy is to set $P_i = \bar{P}_i$ for all channels. The targets are met exactly ($J=0$), and the remaining power budget is left unused (slack).
2. Case B (Insufficient Power): If $P_{tot} < \sum \bar{P}_i$, the power constraint is tight ($\sum P_i = P_{tot}$). The optimal dual variable $\lambda > 0$ must be found to balance the deviations.

D. Closed-Form Solution via Lambert W

For the active set of channels (where $P_i > 0$), the KKT stationarity condition leads to an equation solvable via the Lambert W function ($W_0$). The optimal power for channel $i$ given a dual variable $\lambda$ is:
$$P_i(\lambda) = \frac{2}{\lambda c^2} W_0\left( \frac{\lambda c^2}{2 a_i} 2^{T_i} \right) - \frac{1}{a_i}$$
where $c = \ln 2$. The final allocation is clipped to $[0, \bar{P}_i]$ to enforce the no-overshoot property.

3. Algorithm and Complexity

The authors propose Algorithm 1, a bisection method to find the optimal dual variable $\lambda^*$:

1. Check Regime: If $P_{tot} \geq \sum \bar{P}_i$, return caps $\bar{P}_i$.
2. Bisection: If $P_{tot} < \sum \bar{P}_i$, find $\lambda^*$ such that $\sum P_i(\lambda^*) = P_{tot}$.
   • The total power function $S(\lambda)$ is monotonically decreasing, guaranteeing unique convergence.
   • Complexity: $O(N \log(1/\epsilon))$, where $\epsilon$ is the tolerance.
   • Optimization: Channels with small $a_i T_i$ products are pruned early as $\lambda$ increases, reducing per-iteration cost.

4. Numerical Results and Performance

The paper validates the approach through extensive simulations:

• Accuracy: The closed-form Lambert W solution matches general-purpose numerical solvers (SLSQP) to machine precision.
• Speedup: At $N=1,024$ channels, the proposed algorithm is 1,890× faster than SLSQP (0.011s vs. 21.2s), making it suitable for real-time OFDM systems.
• Target Tracking: Compared to Waterfilling, Uniform Allocation, and Proportional Fairness, the Target-Rate Least-Squares approach achieves significantly lower rate deviations.
  • Example: In Rayleigh fading, the median deviation for Target-Rate was 0.16 bits/s/Hz, compared to 1.18 for Waterfilling.
• Power Efficiency: Unlike Waterfilling, which always spends the full budget (often wasting power on channels that have already exceeded their targets), the proposed method leaves power unused when targets are met, demonstrating superior efficiency for QoS-constrained systems.

5. Significance and Contributions

1. Novel Formulation: It introduces a soft-target tracking objective that bridges the gap between hard-constraint margin adaptation and unconstrained sum-rate maximization.
2. Theoretical Breakthrough: It proves the "no-overshoot" property and establishes convexity on the natural domain, enabling an exact analytical solution.
3. Closed-Form Solution: It provides the first Lambert W-based closed-form solution for least-squares target-rate tracking over parallel channels.
4. Practical Impact: The algorithm's $O(N \log(1/\epsilon))$ complexity and massive speedup over numerical solvers make it highly scalable for modern high-density OFDM/OFDMA systems (e.g., 5G/6G).
5. Differentiability: The smooth quadratic objective makes the solution compatible with gradient-based optimization pipelines and end-to-end learning frameworks, unlike hard-constraint approaches.

Conclusion

This paper presents a fundamental shift in power allocation philosophy: moving from "maximize throughput at all costs" to "meet specific targets efficiently." By leveraging the Lambert W function, the authors provide a mathematically rigorous, computationally efficient, and practically superior method for managing power in systems with strict Quality of Service (QoS) requirements.