The telecommunications industry is currently locked in a race to define the “AI-Native” air interface for 6G. The prevailing vision suggests a complete overhaul of the Physical Layer (PHY), replacing decades of signal processing algorithms with massive, End-to-End (E2E) Neural Networks. While theoretically superior, this approach faces a brutal reality check: physics and power consumption. Running a full neural receiver on a battery-constrained mobile device is currently a computational impossibility.
Nokia Solutions and Networks Oy has filed a patent US12484035B2 that cuts through this hype with a disruptive, pragmatic architecture. Instead of discarding traditional signal processing, they have engineered a hybrid receiver that combines the efficiency of linear algebra with the precision of Machine Learning (ML). This is not just an incremental update; it is a blueprint for how AI will likely be deployed in the real world of wireless communications.
The Problem: The “All-or-Nothing” Trap
Current receiver designs are stuck between two extremes, neither of which is fully sufficient for the demands of future networks.
On one side, we have Traditional Linear Receivers (using algorithms like Minimum Mean Square Error – MMSE). These are computationally efficient and mathematically proven. However, they operate on linear assumptions. In real-world scenarios, signals suffer from non-linear impairments-power amplifier distortion, phase noise, and complex interference patterns-that linear models simply cannot capture, leading to suboptimal performance.
On the other side lies the End-to-End Neural Receiver. This “Black Box” approach uses Deep Learning to process raw antenna signals directly into bits. While it captures non-linearities beautifully, the cost is exorbitant. Training a model to “re-learn” the physics of wave propagation from scratch requires massive neural networks. For a User Equipment (UE) device, the computational load and power drain of such a model are prohibitive.
The industry has been trapped in an “all-or-nothing” mindset: either stick with rigid linear math or switch to heavy, battery-draining AI.
The Breakthrough: A “Split-Brain” Architecture
Nokia’s invention breaks this deadlock by acknowledging a critical truth: we don’t need AI to do everything. We only need it to do what linear math cannot.
The patent introduces a Hybrid Receiver that splits the workload based on competency:
1. The Linear Stage (Deprecoding): The receiver first uses traditional linear algebra for “deprecoding”-specifically, the spatial separation of streams (MIMO detection). Since the channel matrix is generally well-understood, linear equalization is incredibly efficient at reducing the dimensionality of the signal. It simplifies the raw antenna data into estimated symbols.
2. The Non-Linear Stage (ML-Based LLR Generation): These estimated symbols are then fed into a Neural Network. Because the heavy lifting of spatial separation is already done, the input to the neural network is significantly smaller (lower dimension) than raw antenna data. The Neural Network focuses solely on the complex, non-linear task: calculating the Log-Likelihood Ratios (LLRs) for the decoder.
By feeding the Neural Network “pre-digested” data rather than raw signals, Nokia reduces the model size drastically. The AI no longer needs to learn spatial demultiplexing; it only needs to analyze the probability of the bits, correcting for the non-linear noise that the linear stage missed.
Identifying patents like Nokia’s hybrid AI receiver and understanding their technical relevance across wireless standards requires navigating large volumes of global patent and research data.
AI-driven R&D intelligence platforms such as Slate help streamline this process by bringing patents, research papers, and technical documents into a unified workspace, enabling faster, evidence-backed evaluation of emerging wireless and AI technologies.
Market Impact: The Bridge to 6G
This hybrid architecture represents a critical maturation point for AI in the Radio Access Network (RAN). Its impact extends across three key vectors:
Feasibility for Mobile Devices: By significantly reducing the parameter count and computational complexity of the neural network, this architecture makes AI-enhanced reception viable for handsets, not just power-rich base stations. It brings the benefits of Neural Receivers to the edge without destroying battery life.
Performance vs. Efficiency Balance: This approach retains the “super-resolution” capabilities of AI-handling non-linear hardware impairments and complex interference-while maintaining the speed of linear processing. It offers a tangible gain in Block Error Rate (BLER) performance over traditional receivers without the latency penalties of full E2E models.
Strategic IP Positioning: As the industry standardizes 5G Advanced and looks toward 6G, “AI-light” architectures will likely become the standard for initial deployment. Nokia is positioning itself not as a dreamer, but as the architect of the practical AI-RAN.
This patent signals a move away from “AI for the sake of AI” toward “AI for precision.” It validates a future where signal processing is a collaborative effort between deterministic mathematics and probabilistic machine learning.
Looking to uncover breakthrough innovations within patent portfolios?
Write to us to identify high-impact patents and strategic value across portfolios.
