Pulse or digital communications – Receivers – Particular pulse demodulator or detector
Reexamination Certificate
1998-12-30
2002-12-03
Chin, Stephen (Department: 2634)
Pulse or digital communications
Receivers
Particular pulse demodulator or detector
C375S262000, C375S340000, C375S348000, C714S704000, C714S791000
Reexamination Certificate
active
06490327
ABSTRACT:
BACKGROUND OF THE PRESENT INVENTION
1. Field of the Invention
The present invention relates generally to telecommunications systems and methods for improved signal reception, and more particularly to a system and method for improved self-adaptive maximum likelihood sequence detection.
2. Background and Objects of the Present Invention
Cellular telecommunications is one of the fastest growing and most demanding telecommunications applications ever. Today it represents a large and continuously increasing percentage of all new telephone subscriptions around the world. In many cases, cellular solutions successfully compete with traditional wireline networks, such as the Public Switched Telephone Network (PSTN). In the long term, cellular systems, using a digital technology, will become the universal method of telecommunication.
However, one technical difficulty that still needs to be adequately addressed in wireless telecommunications is signal distortion. For example, on top of additive white Gaussian noise (AWGN), a signal to a receiver within a cellular network is subject to multipath fading. A propagation delay is caused by the multiple propagation paths to the receiver due to buildings or terrain. As a result of time delays across the different paths, a succession of discrete pulses representing symbols transmitted across a communications channel are smeared to the point that they are no longer distinguishable as well-defined pulses at the receiving terminal. Instead, the received symbols overlap somewhat causing intersymbol interference (ISI).
With reference now to
FIG. 1
of the drawings, there is illustrated a mathematical model of ISI. This is a filter of order L+1, where ‘L’ is the number of memory elements in the filter and L+1 is the number of filter coefficients (&thgr;
0
, . . . &thgr;
L
). In particular, a symbol stream of discrete symbols s
k
, each a member of a defined and finite alphabet A\, is forwarded on a given channel of a transmitter
10
, which has channel coefficients &thgr; combined therein, which represent the intersymbol interference, to a receiver
14
. Along the transmission route
12
to the receiver
14
, a noise component n
k
is added, represented by adder
16
, which further alters the symbol stream signal into a different symbol stream signal v
k
, which is received by the receiver
14
.
The process of undoing the effects of ISI is referred to as equalization. To assist in equalizing the aforementioned altered symbol stream v
k
received at the receiver
14
, digital telecommunications standards employing Time Division Multiple Access (TDMA) technology, like that of the Global System for Mobile communications (GSM) and the IS-136 standards, employ training or synchronization sequences to facilitate signal demodulation. For example, in GSM, systems employ time slots to transmit a 156.25-bit message, where 22 or more of those bits are utilized as training for channel equalization. It should be understood that synchronization is absolutely necessary in certain circumstances, i.e., to initially align the receiver
14
with the transmitter's
10
signal.
In conventional systems, even after such synchronization occurs, i.e., after establishing timeslot alignment, there is still a need for the training bits in each timeslot. It should be apparent, however, that by minimizing or eliminating the use of these equalization training bits in each time slot, particularly after synchronization has been achieved, this would dramatically increase the information throughput without significant changes to the protocol. In any event, the GSM system utilizes the aforementioned 22 bits as training bits throughout all timeslots of the transmission.
Through the use of the training bits, ISI is corrected in current systems by using the training bits for channel equalization. This is accomplished by ascertaining the aforementioned channel coefficients ⊖ for that communications route or link
12
and modifying (demodulating) the incoming signal in accordance with these coefficients. As a result of the dispersive nature of channels across the time domain due to the noise n
k
, each channel coefficient ⊖ is multiplied with the incoming information, i.e., symbols. It should be understood to those skilled in the art that, with a channel behaving as a filter, the phenomenon of ISI may itself also be modeled as a filter.
The length of the filter is the extent of ISI, and is denoted by ‘L+1’. Accordingly, in a situation where there is no ISI, the channel may be modeled with a single non-zero channel coefficient, where all other coefficients are zero, e.g., 1,0,0. In this example, channel coefficients characterize the behavior of the channel. Assuming L=2, then with ISI, all of the remaining channel coefficients are also non-zero, e.g., 1, 0.5, 0.2.
Accordingly, the receiver
14
must demodulate the incoming symbol stream, v
k
, and overcome both the signal dispersion and the background noise present (AWGN). With the conventional channel equalization technique, both ISI and AWGN are overcome by the use of the aforementioned training sequence bits. As discussed, having knowledge of a transmission result, such as the known training sequence value, one may ascertain what the channel did to the outgoing signal and accordingly demodulate the received signal into the appropriate symbols.
A well-known symbol estimation/detection technique employed to perform maximum likelihood sequence estimation (MLSE) on the input symbol stream is the Viterbi algorithm, which dynamically estimates the most probable sequence of data symbols by maximizing a likelihood function describing the received signal. In general, for a binary system attempting to decipher an N-bit input symbol string, a brute force approach (2
N
) is computationally infeasible, in that it is of exponential order. It should be understood, of course, that for an alphabet of size M, M
N
comparisons are required.
The Viterbi algorithm greatly simplifies this exponential order analysis by focusing on a discrete sequence of candidate symbols stored within L memory elements of the filter, i.e., L symbols in length. As an example, for N=156 bits, representing a time burst's binary data information or symbol content, L may be as low as 2, where bounding the analysis to 4 states (2
2
) is preferred. With the two binary values conventionally represented by +1 and −1, these four states of the filter memory elements are as follows:
States
−1, +1
+1, +1
+1, +1
In the dynamic programming approach described by the Viterbi algorithm, the incremental shortest path for each state is ascertained, maximizing the likelihood function linearly.
The Viterbi algorithm, although currently the technique of choice in the sequence analysis art, primarily for its linear order computational simplicity, nonetheless has its shortcomings. For instance, the Viterbi algorithm requires foreknowledge of the channel and its coefficients, which in the wireless environment are constantly changing due to channel fading, propagation delays and other signal interference conditions. To adaptively equalize the signal from timeslot to timeslot, the Viterbi algorithm therefore requires the aforementioned training bits, which provide the requisite channel coefficient information enabling the receiver to demodulate the signal dynamically.
Although the Viterbi algorithm operates best in situations where the channel coefficients are known, in the case of blind equalization, i.e., where there is no foreknowledge of the channel coefficients, the Viterbi algorithm has proved suboptimal. Thus, the Viterbi algorithm is not well suited for a self-adaptive sequence detection technique, particularly in the blind equalization context, which attempts to detect the transmitted sequence without resorting to a training sequence. Therefore, alternative methodologies are required in such instances.
Such alternative methodologies include tree-search algorithms, such as the M-algorithm, the Fano
Chin Stephen
Ericsson Inc.
Jenkens & Gilchrist P.C.
Liu Shuwang
LandOfFree
System and method for self-adaptive maximum likelihood... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for self-adaptive maximum likelihood..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for self-adaptive maximum likelihood... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2957527