Error detection/correction and fault detection/recovery – Pulse or data error handling – Data pulse evaluation/bit decision
Reexamination Certificate
2001-05-03
2004-02-10
Dildine, R. Stephen (Department: 2133)
Error detection/correction and fault detection/recovery
Pulse or data error handling
Data pulse evaluation/bit decision
C714S780000, C714S794000
Reexamination Certificate
active
06691263
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to decoding in a communications receiver, and, more particularly, to iterative decoders generating reliability information.
2. Description of the Related Art
Digital transmission and recording systems convey discrete-time data sequences across channels using analog signals that vary as they pass through different channel media (e.g., wireless channels or magnetic/optical recording media). A primary goal of any such system is to convey information at the fastest possible rate with a minimum number of errors. Accordingly, numerous approaches for error control have been developed to try to minimize and/or correct errors in transmitted signals, as illustrated in U.S. Pat. No. 6,029,264 to Kobayashi et al. and “Error Control Coding, Fundamentals and Applications,” by S. Lin and D. Costello, Jr., Pages 1-14, Prentice Hall, 1983, both of which are incorporated herein fully by reference.
A data sequence may be defined as comprising, for example, a series of bits represented as analog pulses. The analog pulses either change in amplitude or remain unchanged for each corresponding bit interval, and may either represent the bit itself by an amplitude or represent transitions between bits (e.g., one amplitude represents a transition from “0” to “1”, while a different amplitude may represent a transition from “1” to “0”). The data sequence is thus transmitted as a sequence of analog pulses, each with duration (interval) T. The analog pulse is filtered to select a pulse shape that is optimal in some manner for detection (e.g., via square-root raised cosine filtering). A receiver attempts to reconstruct the data sequence from the received sequence of analog pulses.
Despite the somewhat imprecise correlation between the original data sequence and the transmitted data sequence ultimately input to a receiver, the different pulses may be distinguishable from each other by a detector using a well-known technique referred to as “sequence detection. In contrast to detecting a present, or current, bit as a decision based solely on a sampled, received pulse, sequence detection examines a sequence of samples over several received pulses to detect a present bit. Even if the signal is corrupted by noise, the detection of each bit is accomplished by i) sampling at the bit interval T and ii) storing the sample value of each pulse at the sample point. The bit being sampled at time n is referred to herein as the “present sample” or “present bit” P(n). By comparing the value of the present sample with the value of the samples immediately preceding the present sample in the data sequence it can be determined if a change in amplitude has occurred. The samples immediately preceding the present sample P(n) are the samples taken at time P(n−1), P(n−2), P(n−3), . . . , P(n−M+1), where M is the channel memory. Channel memory M may be considered to be the number of previous samples that influence the present sample. When the present sample is processed, compared, or otherwise under scrutiny, it is also known as the “bit of interest.”
Although sequence detection offers a reasonable method of error control, increasing the overall data rate may cause received pulses to overlap. Pulses may overlap at the point of transmission as each symbol period decreases. Pulses may also overlap due to the channel effect known as dispersion. Since the pulses representing the transmitted bits overlap, the combined pulse, obtained by superposition of the overlapping pulses, results in a signal in which the first pulse and the second pulse are less easily distinguished from one another. This effect is known as inter-symbol interference (ISI). At a receiver, the present sample contains not only the amplitude contribution of the pulse for the bit of interest but also includes the amplitude contribution of the “tail” of one or more previous pulses.
A sequence detector is employed to detect and identify transmitted (or, in the case of magnetic recording systems, recorded) sequences of pulses that are more likely to be incorrectly transmitted, using a priori knowledge of noise characteristics and impulse response of the channel. For example, a sequence detector may use a priori knowledge of the dependence of noise samples on previous noise samples, noise power, and/or partial response polynomials representative of the channel. A common noise model is called Additive White Gaussian Noise (AWGN), and a common channel model is the linear time-invariant channel with memory. In addition, if the pulses have short tails, ISI in the bit of interest is limited to a small number of previous samples, since after some time interval a previous pulse will have little or no effect on the value of the present sample. For typical implementations of sequence detectors, it is desirable that the samples of the signal tail be limited to a small, finite number of integer values to reduce the number of comparisons that must be made. Systems utilizing this property are called “partial response” (PR) systems, and the number of previous samples effecting the pulse of the bit of interest is generally known as the channel memory length.
A maximum-likelihood sequence detector (MLSD) is a sequence detector known in the art and used for an uncoded, linear channel with ISI and AWGN. An MLSD detector comprises a whitened, matched filter (WMF), having an output that is sampled at the Nyquist rate of the input signal, and a Viterbi detector employing a version of the well-known Viterbi algorithm (VA). The sampled output of the WMF is input to the Viterbi detector. The trellis structure of the VA algorithm includes i) a sequence of states and ii) a set of paths through the trellis. The state of the trellis is defined by the ideal values for received bits, and each state typically corresponds to a presently received bit and one or more previously received bits with the number of previously received bits generally equivalent to the length of the channel memory. A path through the trellis is termed an output sequence and defines a sequence of transitions between the states through the trellis, with each path corresponding to a block of possible, or allowed, bits.
The trellis of the VA algorithm describes all possible data sequences that can occur at the output of the channel as a result of the input sequence. A particular path through the trellis is determined from the input samples and reflects the maximum-likelihood data sequence transmitted through the ISI channel. The particular path is determined by comparing the likelihood of traversing each of the paths through the trellis given the input samples. A metric, such as minimum Euclidean distance, between the received samples and ideal symbol values is calculated for each state transition to derive a quantity that reflects a likelihood or transition probability for the input sample making the particular transition to the state (sometimes termed a branch metric). Comparing the likelihood of traversing each of the paths is accomplished by comparing the sum of the branch metrics (sometimes termed the path metric) of each path to determine the path exhibiting the maximum likelihood of occurrence given the input samples. These prior art MLSD detectors are generally not practical, particularly for communication and recording systems. The MLSD detectors are generally not practical because the structure of the MLSD is prohibitively complex to implement when the VA algorithm requires a large number of states in the trellis (i.e., the hardware required to implement all the calculations and comparisons is very complex).
Many PR systems utilize equalization to reduce the residual ISI-caused errors of mis-detection, and possibly reduce the number of states in the trellis used by the VA algorithm. Equalization may typically include amplification of the received signal by an amplifier that has a frequency response based on an estimate of the channel's frequency response. In some PR systems, a data sequence is reconstructed fro
Lee Inkyu
Sonntag Jeffrey L.
Vasic Bane V.
Dildine R. Stephen
Hughes Ian M.
Mendelsohn Steve
LandOfFree
Interative decoding based on dominant error events does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Interative decoding based on dominant error events, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Interative decoding based on dominant error events will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3328006