On-Channel Repeater Adaptive Filter Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional LMS algorithms used in on-channel repeaters face slow convergence and filter estimation errors due to the 'large eigenvalue ratio' problem, leading to spurious components and instability, especially in regions with low signal energy and rapid feedback changes.
Innovation Solution
An additional side-chain with two coefficients is introduced to supplement the LMS estimator, targeting out-of-band regions by filtering and correlating error and reference signals to improve convergence speed and reduce noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the conventional LMS algorithm is used for feedback cancellation in on-channel repeaters, then the system structure remains simple, but the convergence speed is slow and filter estimation errors occur due to the large eigenvalue ratio problem
Solution Approach 1:
The patent segments the frequency spectrum into in-band and out-of-band regions, and applies separate coefficient adaptation mechanisms to each. The LMS algorithm handles in-band frequencies while a separate mechanism handles out-of-band frequencies, allowing optimized convergence for each region without requiring a completely complex new algorithm
Solution Approach 2:
The patent changes the eigenvalue distribution by applying frequency-dependent scaling to the coefficient updates. Different scaling factors are applied to different frequency regions, effectively transforming the eigenvalue ratio problem into a manageable form that enables faster convergence while maintaining algorithmic simplicity
2Measurement precision
If the LMS algorithm operates in regions with low signal energy, then complete frequency coverage is achieved, but filter estimation errors and spurious components increase
Solution Approach 1:
The patent converts the harmful effect of low signal energy in out-of-band regions into a benefit by deliberately allowing these regions to be handled separately with different coefficient adaptation. The low energy regions would otherwise cause estimation errors, but by isolating them and applying appropriate scaling, the system actually improves overall estimation accuracy
Solution Approach 2:
The patent applies different quality characteristics to different frequency regions. In-band regions receive full LMS adaptation with higher precision, while out-of-band regions receive scaled adaptation appropriate to their lower energy levels. This local differentiation prevents spurious components from low-energy regions while maintaining high accuracy in signal-rich regions
3Speed
If the feedback cancellation system uses rapid coefficient adaptation, then convergence speed improves, but numerical overflows and instability occur
Solution Approach 1:
The patent introduces dynamic scaling factors that automatically adjust the coefficient adaptation rate based on the local signal conditions and eigenvalue distribution. This dynamic adjustment allows rapid convergence in well-behaved regions while preventing numerical overflows in sensitive regions, maintaining both speed and stability
Solution Approach 2:
The patent employs a feedback mechanism where the scaled coefficient updates are continuously monitored and adjusted based on their effect on system stability. The scaling factors themselves are adapted based on observed performance, creating a self-regulating system that prevents instability while maintaining fast convergence
4Quantity of substance
If the adaptive filter processes the entire Nyquist band, then complete frequency coverage is achieved, but the large eigenvalue ratio problem persists causing slow convergence
Solution Approach 1:
The patent segments the Nyquist band into in-band and out-of-band regions, processing each with appropriately scaled coefficient adaptation. This segmentation maintains complete frequency coverage while avoiding the large eigenvalue ratio problem by treating different frequency regions differently
Solution Approach 2:
The patent changes the effective eigenvalue distribution across the frequency band by applying frequency-dependent scaling. This parameter transformation maintains full frequency coverage but resolves the eigenvalue ratio issue by equalizing the impact of different frequency regions on convergence
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
An on-channel repeater comprises a receiving antenna (12) for receiving an RF input signal x(t); a transmitting antenna (22) for transmitting a signal on the same frequency as the input signal; and an amplification path (16, 18, 14, 20) between the receiving and transmitting antennas, the amplification path providing substantially linear processing. An input (40) is provided for receiving a reference signal y(t); and means (24) are coupled to the reference signal input for producing a plurality of control coefficients, the coefficients being estimated by use of the least-mean-square algorithm. An adaptive filter (26) is coupled to the reference signal input and controlled by the control coefficients to provide a modified signal. A combiner (16) combines the modified signal with the signal in the amplification path so as to reduce the effect of the feedback. The adaptive filter operates with coefficients h(t) produced by the least-mean-square algorithm combined with an additional set of coefficients hw(t) adapted to improve the speed of convergence of the least-mean-square algorithm. The additional coefficients may be generated by a correlator (74) operating on the filtered (70, 72) reference and output signals, or similar more complex arrangements.