A method and apparatus for signal equalization are provided. Multiple decision components are arranged in a sequence, beginning with a history portion and ending with a decode portion. Each decision component performs a decode decision on a symbol. Decode decisions are passed forward to other decision components where they can be used to compensate for intersymbol interference. Decode decision output by the history portion are otherwise discarded, while decode decisions output by the decode portion are output as a decoded signal. In the next decode cycle, input previously provided to the decode portion is again provided to the history portion, in a sliding, overlapping block manner.

The present disclosure discloses a signal equalization apparatus. A channel length estimation circuit determines a transmission channel length of the input signal such that a processing circuit retrieves predetermined feed-forward equalizer coefficients. A feed-forward equalizer equalizes the input signal according to operation feed-forward equalizer coefficients. An auto gain circuit amplifies the input signal according to an offset signal. A signal adding circuit adds the amplified input signal and a feedback adjusting signal to generate an added input signal. A data slicer generates a data-slicing result and the offset signal according to reference thresholds based on the added input signal. A feedback equalizer equalizes the data-slicing result to generate the feedback adjusting signal according to operation feedback equalizer coefficients. The feed-forward equalizer and the feedback equalizer keeps updating the equalizer coefficients such that a signal interference noise is eliminated rapidly to increase a signal and noise ratio of the system.

Equalization methods and equalizers employing discrete-time filters are provided with dynamic perturbation effect based adaptation. Tap coefficient values may be individually perturbed during the equalization process and the effects on residual ISI monitored to estimate gradient components or rows of a difference matrix. The gradient or difference matrix components may be assembled and filtered to obtain components suitable for calculating tap coefficient updates with reduced adaptation noise. The dynamic perturbation effect based updates may be interpolated with precalculated perturbation effect based updates to enable faster convergence with better accommodation of analog component performance changes attributable to variations in process, supply voltage, and temperature.

An apparatus and a method for handling non-continuous data transfer for a decision feedback equalizer in a memory subsystem. The apparatus includes a plurality of end-of-transfer detection flip-flops configured to sample a read data enable signal; a flag flip-flop; a first logic circuit configured to generate a load enable signal in response to the end-of-transfer detection flip-flops and the flag flip-flop; a second logic circuit configured to generate a load data in response to the end-of-transfer detection flip-flops, the flag flip-flop and the read data enable signal; a plurality of first-in-first-out buffers configured to receive the load enable signal and the load data, and unload the load data as an end-of-transfer indicator in line with data strobe; and a plurality of bypass flip-flops configured to generate a bypass signal in response to the end-of-transfer indicator.

A linear retimer includes an equalizer, a clock recovery circuit, a sample and hold (S/H) circuit, and a linear driver. The equalizer receives an input signal and outputs an equalized signal. The clock recovery circuit receives the equalized signal and outputs a clock signal. The S/H circuit receives the equalized signal and the clock signal and outputs a retimed signal. The linear driver receives the retimed signal and outputs a recovered signal. The S/H circuit is configured to preserve a voltage of the equalized signal in the retimed signal. In some examples, the S/H circuit is part of a linear three-tap feedforward equalizer, and the linear driver receives an output of the feedforward equalizer. The linear retimer can be placed between a transmitter and a channel or after the channel.

Decision feedback equalization (DFE) tap systems and related apparatuses and methods are disclosed. An apparatus includes output nodes to provide output signals, a complementary metal-oxide-semiconductor (CMOS) DFE tap electrically connected to the output nodes, and a current integrating summer electrically connected to the output nodes. The current integrating summer is to reset the output nodes to a common mode voltage potential.

The embodiments described herein provide for a method and system for training an optimal decision feedback equalization (DFE) coefficient for use in GDDR and DDR applications. The method includes determining a first expected bit pattern using a reference voltage. The method further includes determining a transition voltage value of the first expected bit pattern. The method further includes receiving a second expected bit pattern having a same first bit as the first expected bit pattern. The method further includes determining a transition voltage value of the second expected bit pattern using the reference voltage. The method further includes calculating an optimal reference voltage value by averaging the transition voltage values of the first expected bit pattern and the second-expected bit pattern and storing the optimal reference voltage value in a register corresponding to a logic value of the same first bit.

Methods, systems, and apparatus for error correction are provided. In one aspect, an error correction method includes: obtaining an output signal and an amplitude value of a feed forward equalizer (FFE), the amplitude value being a channel response amplitude value corresponding to an equivalent channel of the FFE, performing level decision on the output signal based on the amplitude value to obtain a first decision signal including (2M−1) decision symbols, M being an integer not less than 2, performing (1/(1+D)) decoding on the first decision signal to obtain a first decoded signal, determining a second decision signal based on the first decoded signal, the second decision signal including (M−1) decision symbols, determining that a burst error occurs in the second decision signal if an absolute value of the second decision signal is greater than (M−1), and correcting the burst error in the second decision signal.

A decision feedback equalizer for generating an output signal according to an input signal includes: a feedforward equalizer, a feedback equalizer and a weight coefficient control unit. The feedforward equalizer includes a plurality of tapped delay lines and is controlled by a set of first weight coefficients. The feedback equalizer includes a plurality of tapped delay line and is controlled by a set of second weight coefficients. The weight coefficient control unit is employed to selectively adjust at least one of the set of first weight coefficients and determine a set of first boundary values for at least one of the set of second weight coefficients. When the at least one of the set of second weight coefficients does not exceed the set of first boundary values, the weight coefficient control unit increments the at least one of the set of first weight coefficients.

Pre-charging two or more sets of nodes to set a differential output of a multi-input summation latch connected to the two or more sets of nodes in a pre-charged state, the two or more sets of nodes comprising a set of data signal nodes and a set of DFE correction nodes, in response to a sampling clock, generating a differential data voltage and an aggregate differential DFE correction signal, and generating a data decision by driving the differential output of the multi-input summation latch into one of two possible output states according to a summation of the differential data voltage signal and the aggregate differential DFE correction signal and subsequently holding the data decision by holding the differential output of the multi-input summation latch in a latched state for a duration determined by the sampling clock.