P-53: Automatic LCD Gamma Curve Optimization

P-53 / J. Oh

P-53: Automatic LCD Gamma Curve Optimization Jaeho Oh, Seung-Woo Lee*, Kwan-Young Oh, Taesung Kim, Brian H. Berkeley and Sang Soo Kim LCD Business SAMSUNG ELECTRONICS Co., LTD #200, Myeongam-Ri, Tangjeong-Myeon, Asan-City, Chungcheongnam-Do, Korea(ROK) *also with Dept. of Information Display, Kyung Hee University, Seoul, Korea

Abstract To achieve completely consistent LCD panel-to-panel color performance in mass production, novel algorithms have been developed for automated tuning and programming of LCD gamma. Optimal gamma is derived using a digital gamma buffer (DGB) and a linear scan method (LSM) as a simple but accurate solution. Another approach uses a V-T curve interpolation method to achieve shorter tact time.

1.

Introduction

Significant research has been conducted on gamma curve modification for improved color performance [1-3]. The luminance curve or gamma curve is defined using the gamma value as described in Eq. (1).

Luminance

 Gray  =   255 

gamma

× Maximum

luminance

The design point of the LCD panel in Fig. 1 is gamma 2.2, however, the center value of the distribution is shifted to a higher value. To control the gamma curve, the simplest method is to adjust gamma reference voltages. Conventionally, gamma reference voltages are generated by a simple resistor string as shown in Fig. 2. However, with fixed resistor values, it is not possible to adjust the reference voltages. Adjustable and programmable gamma reference values are needed to compensate for the wide spread in panel-to-panel gamma variation. For this purpose, we have developed a new programmable digital gamma buffer (DGB) as shown in Fig.3. This paper proposes a new algorithm for fast and effective gamma optimization using the DGB. This buffer is a key enabling technology for Samsung’s advanced S-PVA technology [3]. VREFUH

Eq. (1)

O_REF_UH

Bank Bank B

Gamma is one of the most important optical performance characteristics of a display. The gamma curve of an LCD is affected by the liquid crystal material, the cell gap and pixel structure, reference voltages, and other factors. Even though these factors may be fixed from a design standpoint, gamma values of mass-manufactured LCD panels have a very wide distribution from the target gamma due to process variations. Fig.1 shows a typical spread of gamma values for one mass-produced model.

SDA

2

IC Control

SCL

A

DAC High

OUT1 OUT2 OUT3 OUT4 OUT5 OUT6 OUT7 O_REF_UL

VREFUL BANK_SEL VREFLH

O_REF_LH

Bank Bank

A

B

DAC Low

OUT8 OUT9 OUT10 OUT11 OUT12 OUT13 OUT14 O_REF_LL

VREFLL

Figure 3. Digital Gamma Buffer

2.

Gamma Figure 1. Typical LCD panel gamma variation Reference voltages

Automatic Gamma Optimization

Fig. 3 shows that the DGB has 18 reference voltage outputs, including 14 digitally programmable channels (OUT1 – OUT14), each with 8-bit resolution. The 8-bit digital values stored in the registers, banks A and B, correspond to values for the S-PVA A and B sub-pixels. These values are converted to analog voltages by the linear DACs. Therefore, there are 7 programmable reference outputs or tap points per polarity, distributed as shown in Fig. 4. Gamma variation is caused by variation in the relationship between the voltage (V) applied to the liquid crystal and its transmittance (T) characteristics. This is to say that every panel has its own unique V-T characteristics. Therefore, the actual gamma curve deviates from the target curve as shown in Fig. 5.

Figure 2. Gamma reference voltage generation

394 • SID 06 DIGEST

ISSN0006-0966X/06/3701-0394-$1.00+.00 © 2006 SID

P-53 / J. Oh Luminance, L

1.0 0.9

Normalized Luminance

0.8 0.7

L2 LTarget

0.6

L1

0.5 0.4 0.3 0.2 0.1

D1

0.0 0

32

64

96 128 160 Input Gray Level

192

224

256

Figure 6. Linear Scan Method Thus, we can express the linear relation as shown in Eq. (2), where a and b are coefficients. If we measure luminance (L1 and L2) according to two different digital trial settings, D1 and D2, respectively, Eq. (2) can be modified as Eq. (3). From Eq. (1), target luminance (Ltarget) can be calculated once target gamma is given. Thus, the target digital value (Dtarget) can be obtained by rearranging Eq. (3) into Eq. (4).

Measured curve Target curve

0.9

Normalized Luminance

0.8 0.7 0.6 0.5 0.4

Eq. (2)

L = aD + b

0.3

L =

0.2 0.1 0.0 0

32

64

96 128 160 Input Gray Level

192

224

256

Figure 5. Actual gamma curve vs. target gamma curve We can fit the measured gamma curve to the target curve by adjusting the applied voltages of the programmable tap points. Next, we will describe a method to find the optimal reference voltages, i.e. digital gamma values for each tap point.

2.1

D2

Digital gamma data, D

Figure 4. 7 programmable tap points vs. input image gray level 1.0

DTarget

Linear Scan Method (LSM)

We can make the assumption that within a short range, the luminance is linearly proportional to the applied reference voltages as shown in Fig. 6. Because we are using a linear DAC in the DGB, the applied voltage is directly expressed by the digital values as shown in Fig. 6. The linear scan method is an algorithm which finds optimal digital values of the tap points to meet the target gamma curve based on the assumption of a linear relationship between luminance (L) and digital gamma data (D) at each tap point.

( L 2 − L 1 )( D − D 1 ) + L1 (D 2 − D1)

D target

=

( L target − L 1 )( D 2 − D 1 ) ( L 2 − L1 )

Eq. (3)

+ D1

Eq. (4)

The derived Dtarget using linear approximation is not exactly the same as the optimal value, DX, due to nonlinearity of the V-T curve. However, the error (Dtarget - DX.) is small. Experimentally, we have found Dtarget will be within the range of DX ± 3. We can find an exact match to the optimized value using additional iterations. If the gamma spread is less than 1, for example 1.7 < gamma < 2.7 when the target gamma is 2.2, then LSM is an effective method to adjust gamma reference voltages. If the gamma spread happens to be greater than 1, we could again use a linear approximation to describe the relationship between voltage and luminance, however, more iterations would be required, which would increase gamma tuning time. Naturally, a higher order polynomial could be used for the approximation in order to reduce the number of required iterations. However, a higher order equation would require more samples, i.e., additional measurement time. Therefore, short measurement time is an advantage for the LSM. But when the gamma spread is greater than 1, or when more than two gammas are required such as for S-PVA [3], the performance of LSM is not satisfactory. In that case, a more effective interpolation method is needed.

SID 06 DIGEST • 395

P-53 / J. Oh 2.2

Interpolation method using V-T Curve

An LCD panel’s V-T curve describes the relationship between transmittance and applied voltage. Once the V-T curve has been measured, the gamma curve can be extracted. Measurement of the V-T curve is time consuming because all luminance values must be measured for every voltage point. In case of an 8-bit digital gamma buffer, 256 measurements are required. Due to slow response time of the liquid crystal, at least 150ms are needed per data capture to allow the LC adequate time to settle before each measurement. Therefore, about 40 seconds are required to collect enough luminance data just for one V-T curve. Considering processing time on the manufacturing line, it would take too long to measure each individual point. Therefore, we propose an interpolation method to generate the V-T curve based on a minimum number of samples. Compared to the LSM, the interpolation method is very effective, particularly for S-PVA.

Table 1. Error Factor table Equation Order Sampled data

1st

2nd

3rd

12

1.82

1.28

0.75

17

1.47

0.95

0.61

23

0.77

0.51

0.44

27

0.65

0.42

0.43

33

0.44

0.26

0.20

600

Luminance [cd/m 2]

500 400 300 200

Figure 8. Error factor graph

100 0 0

32

64

96

128 160 192 224 256

Digital gamma buffer data, D

Figure 7. Concept of the interpolation method Fig 7 shows the concept of the proposed V-T curve interpolation method. The interpolation intervals are allocated according to the instantaneous slope of V-T curve. The dark portion of the curve changes more rapidly. As a result, the low luminance region requires more samples for accurate curve fitting compared to the high brightness region. Experimentally, we found that more samples are needed below gray level 112. It is important to determine the number of samples and the order of polynomial equation required for interpolation. To evaluate the level of deviation, an error factor is proposed as follows: n

ErrorFactor =

∑

( Lmeasure − Lcalculation ) 2

0

N

Eq. (5)

where n = 0, 1, …, N; and N = 256 for an 8 bit panel This error factor is a summation of root mean square differences between Lmeasure and Lcalculation, where Lmeasure is measured luminance and Lcalculation is calculated luminance using the interpolation method. We used this error factor to assess effectiveness of the interpolation. Table 1 shows that the experimental error factor results depend on the number of samples and the order of equation used for interpolation. Fig. 8 is a graph of Table 1.

396 • SID 06 DIGEST

Acceptable gamma deviation level needs to be identified in order to find the optimal condition. We found that with an error factor of 0.42, it is possible to achieve a gamma spread of