Prior Art Discussion Widely available systems to detect data on optical storage media illuminate a"spot" on the medium, and mutual movement causes the spot to travel along a track of data- bearing pits (or"symbols"). The typical approach is to centre the spot at any one time on a pit in order to optimise imaging of the pits.

A problem which has presented itself with prior approaches is that noise in the detected signal is introduced by neighbouring pits which are captured or partially captured in the image for a spot position centred on the pit of interest. The approach to minimising such noise has been to introduce guard bands around the tracks. These however occupy a certain proportion of the medium's surface area and thus reduce data density per unit area. Heretofore, this limitation has been compensated for by virtue of improvements in laser emitters and detectors: allowing ever decreasing spot, and hence pit, sizes. However, as limitations on miniaturisation of the optics are reached, the loss of data density caused by use of guard bands is more noticeable.

The invention is directed towards addressing this problem.

SUMMARY OF THE INVENTION According to the invention, there is provided a data detector system for detecting data on an optical storage medium, the system comprising:

an optical pickup unit for capturing a plurality of samples, each sample being captured for an illuminated area which is offset from symbols of the medium whereby each sample represents data from multiple symbols; and a data detector for processing said plurality of samples to provide data decisions.

In one embodiment, the illuminated area is centred on a centroid between neighbouring symbols, the centroid being a uniform distance from the symbols.

In another embodiment, the symbols have a hexagonal lattice pattern, and the illuminated area is centred on a centroid between each successive group of three symbols in a triangular pattern.

In a further embodiment, the illuminated area encompasses the three symbols close to the edge of the illuminated area.

In one embodiment, the symbols have a rectangular lattice pattern, and the illuminated area is centred on a centroid between each successive group of four symbols.

In another embodiment, each group of four symbols has a square pattern.

In a further embodiment, the data detector executes a sum-product algorithm by considering each sample of an equalized signal as a constraint that needs to be satisfied.

In one embodiment, the samples are stored as static data, and values of symbol estimations are dynamically updated in iterative cycles.

In another embodiment, each symbol estimation is a likelihood value arising from sum-product processing.

In a further embodiment, the iterative cycles end when the sample estimations converge.

In one embodiment, the data detector comprises a plurality of processors operating in parallel, each generating likelihood information based on a previous likelihood information and the plurality of samples, and passing its generated likelihood information to a next processor.

In a further embodiment, a final processor makes a final hard symbol decision.

In another aspect, the invention provides a method of storing data on an optical storage medium and of reading said data, the method comprising the steps of : writing symbols to an active surface of the medium in a pattern having tracks in parallel without guard bands therebetween; an optical pickup successively illuminating areas, each centred on a centroid between a group of neighbouring symbols, and capturing a sample for each area; and a data detector estimating symbol values by treating each sample as a constraint that needs to be satisfied.

In one embodiment, the symbols are written to the active surface in a hexagonal pattern, and each illuminated area is centred on a centroid between a group of three symbols in a triangular pattern.

In another embodiment, the data detector uses a sum-product algorithm for symbol estimation in iterative updates of symbol values until convergence.

DETAILED DESCRIPTION OF THE INVENTION Brief Description of the Drawings The invention will be more clearly understood from the following description of some embodiments thereof, given by way of example only with reference to the accompanying drawings in which:- Fig. 1 is a general plan view of an optical data storage disk; Figs. 2 and 3 are diagrams showing spot location for readings in rectangular and hexagonal pit patterns; Fig. 4 is a plot illustrating system performance; Figs. 5 and 6 are diagrams showing a constraint at each signal sample for sum-product data processing, and flow of likelihood information to bit arC respectively; and Fig. 7 is a plot illustrating simulated system performance.

Description of the Embodiments Referring to Fig. 1 an optical storage disc 1 has a spiral track pattern 2, each meta track having the NR rows of pits (symbols). Within each meta track there are no guard bands between rows of pits, however, there is a guard band between meta tracks. Thus, the pattern 2 allows a data density improvement of 20% to 30% compared to patterns having a guard band between every row of pits. This, however, introduces greater inter-symbol interference (ISI). The problem of ISI is avoided by the manner in which a data detection system of the invention operates.

Referring to Fig. 2, there are pits 10 in a generally rectangular pattern: extending in orthogonal dimensions. A laser spot is indicated by the circle 11 at a position for a

reading. As is clear from this diagram the spot 11 is not centred on a pit. Instead, its centre is at the furthest location from a pit, namely the centroid between a square formed from four pits. The direction of spot"travel"is indicated by the arrow 12.

Referring to Fig. 3, an array of pits 20 has what is referred to as a"hexagonal" pattern, one dimension extending at 45° to the other. In this case a laser spot 21 is centred at the centroid of a triangle formed by three pits 20. The direction of travel is indicated by the arrow 22.

A data detector system of the invention comprises an optical pickup and a data detector. The optical pickup illuminates an area ("spot") of the medium. A sample is captured for each spot position. The data detector processes the readings.

Taking the embodiment of Fig. 3, the optical pickup captures a sample for each successive spot position. At a certain point in time all of the samples for a disc area are processed simultaneously. The processing uses the sum-product algorithm, and in simple terms it involves the following steps: - If a value of close to 3 is detected, it treats all three symbols as bit 1.

- If the value is close to zero it treats all three symbols as bit 0.

- If the value is close to 2 or 1, it takes into account the values for the adjacent readings. It can use a"process of elimination"to estimate where the zero value symbol is according to values of the neighbouring readings.

Thus, by taking samples for spot positions offset from physical pit (symbol) positions, and using computational estimation techniques the detector can read the data even though there are no guard bands between adjacent pit rows.

In more detail, the lattice patterns of Figs. 2 and 3 and other alternatives may be described in more detail as follows.

The data is sampled with the same lattice as the data is written to the medium, which is effectively baud rate sampling for the two dimensional array and hence a low pass filter is employed to avoid aliasing. The nature of this anti-aliasing low pass filter is related to the sampling lattice employed. Both a rectangular (square) lattice (Fig. 2) and an hexagonal lattice are (Fig. 3) are considered. Any two dimensional lattice may be described by the sample locations:

where V is the sampling matrix i and j being integer indexes and i E {O, 1,..., NR-1} andj E {...,-1, 0,1, 2,...}. For a square lattice dx = 1 and dy = 1, with 1 being the lattice parameter and the sampling matrix: For hexagonal sampling in a 2-dimensional plane, the sampling matrix may be written as: For proper hexagonal sampling, In each case, the low-pass filter HLPF (#x,#y) needs to be chosen to cover a fundamental period of the spatial frequency plane. Thus, a low-pass filter with f1=1/2l and f2=1/2l is required for the square lattice and the low-pass filter with f1=1/2l and f2 = 1/#3l is required in hexagonal lattice.

Each sample has an associated area Idet (V) § and hence the storage density on the <BR> <BR> <BR> medium is p = 1<BR> #det(v)## The hexagonal pattern achieves the higher density.

For operation of the data detector, maximum likelihood detection can be achieved using the Viterbi algorithm. This requires representing the channel as a finite state

machine. However, due to the two-dimensional nature of the channel, the state representation of such a machine will consist of columns of input data, thus rapidly increasing the complexity of a complete Viterbi detector. Even in the case of modest ISI, the number of states required in a Viterbi detector would be 2 NRR which would be in excess of one million for NR being a practical value of 10. In practice, the optical spot can extend further thus rendering the Viterbi detector impractical to implement. Thus, while the Viterbi algorithm may be used by the data detector, it is likely to be too complex for most applications.

The response cannot be decomposed into horizontal and vertical components easily.

However, good detection performance can be achieved by application of the sum- product algorithm. This can be applied by considering each sample of the equalized signal as a constraint that needs to be satisfied. Fig. 5 shows the basic constraint for each received signal sample. The constraint for the sample at row r and column c is Yr,c = ar,c + ar,c+1 + ar+1,c + nr,c where Yr,c is the received reading or sample value, a,, c is the bit value E {0, 1} and , e is the noise on the received sample value.

The data detector operates by loading the Yr,c values to memory as the static, sample values. Initial, default, symbol values ar,c are also written to a memory array. The data detector then uses the sum-product algorithm to generate a fresh set of values for arC. This cycle is repeated, with the detector iteratively updating the arC values, until they stabilise and converge.

Assume the log likelihoods of bit ar, c are (the likelihoods of logic 1 and logic 0 are shown explicitly here but in practice can be represented as a single value) Lo (ar,c) α log@= (ar, c = 0) and Ll (ar,c) α log(p=(ar,c = 1) ). The detector proceeds by iteratively updating the likelihoods.

Initially all bits are set to be unknown i. e. L k (a,,,) =0 = L 1k(ar,c) for all c and 0 < r < NR-1. For r < 0 and r # NR, the bits are known to be 0 in the guard band at both sides of the meta-track and thus Lk (a,,,) = 0 and L k (ar,c) = - #.

They are then updated iteratively as follows: Likelihood information for a,., comes from ar,c+1 and ar+1,c using sample yr,c. This <BR> <BR> <BR> update information can calculated by using the two functions Fo () and Fl () which are defined, using logMAX type arithmetic as Fl ,c+1,ar+1,c,yr,c) = max ( L0k(ar,c+1) + L0k(ar+1,c) = (yr,c-1.0)2, L0k(ar,c+1) + L1k(ar+1,c) - (yr,c-2.0)2, L1k(ar,c+1) + L0k(ar+1,c) - (yr,c-2.0)2, L1k(ar,c+1) + L1k(ar+1,c) - (yr,c-3.0)2 ) and and max ( L0k(ar,c+1) + L0k(ar+1,c) - (yr,c-0.0)2, L0k(a,rc+1) + L1k(ar+1,c) - (yr,c-1.0)2, L1k(ar,c+1) + L0k(ar+1,c) - (yr,c-1.0)2, L1k(ar,c+1) + L1k(ar+1,c) - (yr,c-2.0)2 ) However, likelihood information for a,,, also comes from ar l, ^ and -i, c+i using sample yr-1,c and also from ar, c-1 and ar+1,c-1 using sample yr,c-1, Thus the complete iteration update for ar, can be calculated as L1k+1(ar,c) = L1k+1(ar,c) +Fl (ar,c+1,ar+1,c,yr,c) +Fl (ar- 1 ,c,ar-1,c+1,yr-1,c) +Fl (ar, c-1,ar+1,c-1,yr,c-1) and <BR> <BR> <BR> <BR> L O (ar,c) = L0k+1 (ar,c)<BR> <BR> <BR> <BR> <BR> <BR> +F0(ar,c+1,ar+1,c,yr,c) +F0(ar-1,c,ar-1,c+1,yr-1,c) + Fo c-1,ar+1,c-1,yr,c-1)

This flow of likelihood information to bit a,., is shown graphically in Fig. 6. This is done in a parallel fashion for all bits. Final decisions are based on comparing L ok (a) to L i (a) yielding a hard decision. Clearly, there are cycles in the graph representing the constraints due to the tribinary partial response and hence the optimum solution via the sum-product algorithm cannot be guaranteed. However, simulation shows that the algorithm performs well. Fig. 7 shows the simulated performance of the full tribinary Viterbi detector (with 7 rows) the sum-product algorithm with 8 iterations and 7 rows and the theoretical performance based on a minimum distance of 3 i. e.

Q (l2a-). The SNR is arbitrarily defined as 10 logio (1. 0lo2) for this plot.

The performance of processing using the sum-product algorithm is very close to the ideal performance and thus allows a practical method of implementing a detector for large row sizes with a complexity per bit that is relatively independent of the row size. A particularly fast response is achieved if the data detector comprises a plurality of processors operating in parallel, each generating likelihood information based on a previous likelihood information and the plurality of samples, and passing its generated likelihood information onto the next processor. The final processor produces the final hard symbol decision.

In practice a sliding window detector is preferable and Fig. 7 also shows the simulated performance of such a detector using a block length of 15 with an overlap of 5 between blocks (i. e. 15-2*5 = 5 decision columns per block). With 8 iterations, essentially optimum performance is achieved. The complexity of the algorithm is linear in the number of rows NR as opposed to the full two dimensional Viterbi detector whose complexity is exponential in the number of rows NR.

It will therefore be appreciated that the invention achieves effective reading of symbols in an exceptionally high density symbol pattern, without guard bands.

The invention is not limited to the embodiments described but may be varied in construction and detail.