Digital Global Systems Inc v. DeepSig Inc

1. Case Identification

• Case #: IPR2024-01358
• Patent #: 11,018,704
• Filed: August 28, 2024
• Petitioner(s): Digital Global Systems, Inc.
• Patent Owner(s): DeepSig, Inc.
• Challenged Claims: 1-24

2. Patent Overview

• Title: Correcting Distortion of Radio Signals
• Brief Description: The ’704 patent discloses techniques for mitigating radio signal distortion in cellular communication systems. The invention uses a nonlinear pre-distortion machine learning model (NPDMLM), such as a neural network, to generate a pre-distorted signal that counteracts distortion introduced by components like power amplifiers.

3. Grounds for Unpatentability

Ground 1: Claims 1-3, 7-11, 15-19, and 23-24 are obvious over Jüschke.

• Prior Art Relied Upon: Jüschke (EP 2,538,553).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued Jüschke disclosed all limitations of the challenged claims. Jüschke taught a "pre-distortion neural network processor" for mitigating transmitter impairments in wireless communication systems. This processor, a type of nonlinear machine learning model, was disclosed for use in mobile communication systems (e.g., LTE) and could be implemented in base stations. Jüschke's model used "weight and/or bias parameters" that were optimized for specific deployment scenarios, such as an "in-line" configuration where the pre-distortion unit precedes a power amplifier, directly mapping to the patent's claims. Jüschke's system generated a pre-distorted signal by processing an input signal, which was then sent through a transmitting stage (including an amplifier) to produce a final transmit output signal for radio receivers.
  • Motivation to Combine (for §103 grounds): N/A (single reference ground).
  • Expectation of Success (for §103 grounds): N/A (single reference ground).

Ground 2: Claims 1-24 are obvious over Jüschke in view of Holt.

• Prior Art Relied Upon: Jüschke (EP 2,538,553) and Holt (Patent 10,552,738).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground asserted that to the extent Jüschke did not teach certain limitations, particularly those in claims 4-6 related to training the model, Holt supplied the missing elements. Holt taught training machine learning models for communication systems using a "loss function" that described the difference between data input to an encoder (transmitter) and data received from a decoder (receiver). Petitioner argued Holt’s method of calculating a "distance metric" (via its loss function) based on an actually received signal and using it to update model parameters was directly applicable to Jüschke's system. The combination taught obtaining a received radio signal, determining a distance metric between it and the original transmit signal, and updating the model parameters based on that metric, as claimed.
  • Motivation to Combine (for §103 grounds): A POSITA would combine Jüschke and Holt to improve the accuracy of Jüschke’s training model. Holt's use of a loss function based on signals recovered by a receiver provided a more accurate training dataset that accounted for real-world transmission and reception effects. This represented a simple substitution of one known training technique (Jüschke's internal feedback loop) for another known, more robust technique (Holt's end-to-end training method) to achieve a predictable improvement in performance.
  • Expectation of Success (for §103 grounds): A POSITA would have a reasonable expectation of success because both references operated in the same technical field of using neural networks in wireless communications and addressed similar problems. Implementing Holt's training techniques into Jüschke's hardware architecture was a straightforward application of known methods.

Ground 3: Claims 2, 10, 18 and their dependents are obvious over Jüschke in view of Holt and Dzierwa.

• Prior Art Relied Upon: Jüschke (EP 2,538,553), Holt (Patent 10,552,738), and Dzierwa (Patent 10,122,479).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground argued that if the combination of Jüschke and Holt failed to render certain claims obvious (specifically claim 2 and its dependents concerning generating a pre-distorted signal for a "particular deployment scenario"), the addition of Dzierwa would do so. Dzierwa taught a machine learning system that observed and learned an RF environment by generating a "knowledge map" based on learning data. It used an "optimization module" to gather environmental parameters (e.g., noise, antenna hardware parameters) to optimize a signal of interest. Petitioner contended Dzierwa's teachings on using environmental data to optimize predistortion for a particular scenario could be directly integrated into the Jüschke/Holt framework.
  • Motivation to Combine (for §103 grounds): A POSITA would be motivated to incorporate Dzierwa's environmental learning capabilities into the Jüschke/Holt system to further enhance performance. By adding Dzierwa's optimization module, the predistortion model would become more adaptive and robust, capable of optimizing its parameters based on real-time environmental conditions. This combination of familiar elements from the same technical field would have provided a predictable improvement in the model's ability to correct distortion in specific, dynamic deployment scenarios.
  • Expectation of Success (for §103 grounds): A POSITA would have an expectation of success in combining these references, as it involved applying Dzierwa's known technique for environmental optimization to the known predistortion system of Jüschke and Holt. The integration would predictably improve performance by providing more complete and relevant data for model training.

4. Arguments Regarding Discretionary Denial

• Petitioner argued that discretionary denial under Fintiv would be inappropriate. Factors 1-5 favored institution because the ’704 patent was not subject to any parallel district court litigation. Factor 6 also favored institution because the petition presented strong, well-supported grounds for invalidity.
• Petitioner further argued that denial under §325(d) based on Advanced Bionics factors would be inappropriate because none of the asserted prior art references (Jüschke, Holt, or Dzierwa) were previously applied during the prosecution of the ’704 patent, nor were they cumulative to the art of record.

5. Relief Requested

• Petitioner requested the institution of an inter partes review and the cancellation of claims 1-24 of the ’704 patent as unpatentable.