71 results about "Log domain" patented technology

Servo control using ferromagnetic core material and electrical windings is based on monitoring of winding currents and voltages and inference of magnetic flux, a force indication; and magnetic gap, a position indication. Third order nonlinear servo control is split into nested control loops: a fast nonlinear first-order inner loop causing flux to track a target by varying a voltage output; and a slower almost linear second-order outer loop causing magnetic gap to track a target by controlling the flux target of the inner loop. The inner loop uses efficient switching regulation, preferably based on controlled feedback instabilities, to control voltage output. The outer loop achieves damping and accurate convergence using proportional, time-integral, and time-derivative gain terms. The time-integral feedback may be based on measured and target solenoid drive currents, adjusting the magnetic gap for force balance at the target current. Incorporation of permanent magnet material permits the target current to be zero, achieving levitation with low power, including for a monorail deriving propulsion from the levitation magnets. Linear magnetic approximations lead to the simplest controller, but nonlinear analog computation in the log domain yields a better controller with relatively few parts. When servo-controlled solenoids provide actuation of a pump piston and valves, electronic LC resonance measurements determine liquid volume and gas bubble volume.

Decoding LDPC (Low Density Parity Check) code and graphs using multiplication (or addition in log-domain) on both sides of bipartite graph. A means for decoding LDPC coded signals is presented whereby edge messages may be updated using only multiplication (or log domain addition). By appropriate modification of the various calculations that need to be performed when updating edge messages, the calculations may be reduced to only performing product of terms functions. When implementing such functionality in hardware within a communication device that is operable to decode LDPC coded signals, this reduction in processing complexity greatly eases the actual hardware's complexity as well. A significant savings in processing resources, memory, memory management concerns, and other performance driving parameters may be made.

A method of processing a layer in a convolutional neural network includes reading a plurality of different subsets of pairs of input vector values and corresponding weight vector values from an on-chip memory, storing the pair values in registers of a plurality of multiplication units, simultaneously processing the different subsets of pairs of values in the registers by multiplying the pairs of values and accumulating the results of the multiplications to obtain an output vector for a current layer. The input vector and the weight vector have floating point values and multiplying the pairs of values comprises adding corresponding pairs of integers in a log domain, and an input value or weight value is released from the on-chip memory after said input value or weight value is no longer needed for calculating non-computed output vector values.