Hardware matrix computation for wireless receivers
In one embodiment, a receiver including one or more signal-processing blocks and a hardware-based matrix co-processor. The one or more signal-processing blocks are adapted to generate a processed signal from a received signal. The hardware-based matrix co-processor includes two or more different matrix-computation engines, each adapted to perform a different matrix computation, and one or more shared hardware-computation units, each adapted to perform a mathematical operation. At least one signal-processing block is adapted to offload matrix-based signal processing to the hardware-based matrix co-processor. Each of the two or more different matrix-computation engines is adapted to offload the same type of mathematical processing to at least one of the one or more shared hardware-computation units.
1. A receiver comprising:
one or more signal-processing blocks adapted to generate a processed signal from a received signal; and
a hardware-based matrix co-processor comprising:
two or more different matrix-computation engines, each adapted to perform a different matrix computation; and
one or more shared hardware-computation units, each adapted to perform a mathematical operation, wherein:
at least one signal-processing block is adapted to offload matrix-based signal processing to the hardware-based matrix co-processor; and
each of the two or more different matrix-computation engines is adapted to offload the same type of mathematical processing to at least one of the one or more shared hardware-computation units.
2. The invention of claim 1 , further comprising one or more memories.
3. The invention of claim 2 , wherein the one or more memories comprise a block active register for storing an indicator of whether the matrix co-processor is in the process of performing a matrix computation.
4. The invention of claim 2 , wherein the one or more memories comprise one or more matrix registers for storing elements of a matrix in connection with in a matrix computation.
5. The invention of claim 2 , wherein the one or more memories comprise one or more channel-coefficient registers for storing channel coefficients in connection with a matrix computation.
6. The invention of claim 2 , wherein the one or more memories comprise one or more channel-weight registers for storing channel weights in connection with a matrix computation.
7. The invention of claim 1 , wherein the one or more shared hardware-computation units comprise at least one of a multiplier, an inverse square root computer, and a subtractor.
8. The invention of claim 1 , wherein the two or more different matrix-computation engines comprise a matrix-decomposition engine.
9. The invention of claim 8 , further comprising one or more memories, wherein:
the one or more memories comprise one or more matrix registers for storing elements of an original matrix for decomposition by the matrix-decomposition engine; and
the elements of a decomposed matrix provided by the matrix-decomposition engine are stored in the same memory locations of the matrix registers at which the original matrix were previously stored.
10. The invention of claim 1 , wherein the two or more different matrix-computation engines comprise a matrix-solution engine.
11. The invention of claim 10 , further comprising:
one or more memories, wherein the one or more memories comprise:
one or more matrix registers for storing elements of an original matrix to be solved by the matrix-solution engine;
one or more channel-coefficient registers for storing channel coefficients provided to the matrix-solution engine for solving the original matrix; and
one or more channel-weight registers for storing channel weights provided by the matrix-solution engine as a result of solving the original matrix.
12. The invention of claim 1 , wherein the receiver is a rake-based or equalizer-based Linear Minimum-Mean Square-Error (LMMSE) receiver.
13. The invention of claim 1 , wherein the at least one signal-processing block is a chip-pulse matched filter, a Linear Minimum-Mean Square-Error (LMMSE) filter, or a demodulator.
14. The invention of claim 1 , further comprising an address and/or data bus coupling at least one of the signal-processing blocks to at least one of (i) another signal-processing block and (ii) the hardware-based matrix co-processor.
15. The invention of claim 1 , further comprising a programmable processing core adapted to implement one or more of the signal-processing blocks.
16. The invention of claim 1 , further comprising a programmable processing core adapted to implement a dedicated hardware datapath coupling at least one of the signal-processing blocks to at least one of (i) another signal-processing block and (ii) the hardware-based matrix co-processor.
17. A method of processing a received signal comprising:
generating a processed signal from a received signal using one or more signal-processing blocks adapted to implement a signal-processing operation;
wherein:
the signal-processing operation involves a matrix computation;
a hardware-based matrix co-processor is used for the matrix computation;
the matrix computation involves at least one mathematical operation;
one or more hardware-computation units of the matrix co-processor are used to perform the mathematical operation; and
the matrix computation offloads the same type of mathematical processing to at least one of the one or more hardware-computation units.