Abstract— The DVB-T2 standard is widely used in recent years, which is based on orthogonal frequency division multiplexing (OFDM) systems. In OFDM system, frequency-domain pilots are usually employed to perform channel estimation. To reduce the number of pilots and hence improve spectral efficiency, decision-directed channel estimation (DDCE) has been considered in the new DVB-T2 standard. In this paper we will address the noise amplification problem of DDCE in DVB-T2 systems. We propose to employ the weighted least-squares method to improve the performance of DDCE in DVB-T2 systems. Simulation results show that the proposed scheme considerably improves the mean-squared estimation errors as compared with existing algorithms in [2,3]. I. INTRODUCTION In OFDM-based systems, pilot symbols are commonly used to estimate the channel and perform channel equalization. However, pilot symbols contribute to overhead to reduce spectral efficiency. The DVB-T2 standard [1] offers a special pilot pattern in which the pilot overhead is minimized. Using this special pilot pattern(PP8), the receiver has to employ additional means to estimate the channel. The PP8 scheme of DVB-T2 was mainly designed to use decision data as pilots. Provided that the data can be correctly detected, each subcarrier channel response can be estimated by dividing the received data by the detected data. However, if higher-order modulation is employed to generate transmitted data, subcarriers with data at inner constellation points will suffer from larger channel estimation errors due to the noise amplification effect. In order to improve the performance of decision directed channel estimation in DVB-T2 systems, [3] proposed to use only outer constellation points for channel estimation. Channel coefficients of subcarriers at inner constellation points will then be obtained by linear interpolation from neighboring subcarriers. Thus, a pre-defined threshold will be used for the classification of inner and outer constellation points. However, the performance of [3] will face the tradeoff between interpolation error and noise effect. If a larger threshold is used, the performance will be dominated by interpolation errors since channel estimation is performed on a smaller number of subcarriers and on the other hand, the use of a smaller threshold to reduce interpolation error will result in classifying most of the data as outer constellation points, without effectively solving the noise amplification problem. In this paper, we propose to apply the weighted lease- square method [4] to improve the performance of decision- directed channel estimation in the DVB-T2 system. By setting the diagonal weighting matrix with elements proportional with the squared magnitudes of the constellation points and solving for the channel impulse response, information of all received data will be used in computing the channel estimate. The complete channel frequency response at each subcarrier can then be obtained by computing the FFT of the impulse response vector. Improved performance will thus be obtained by the proposed method. II. PROPOSED METHOD The received signal of an OFDM system can be described as (, ) (, ) (, ) (, ) ynk xnk H nk wnk    , (1) where ( , ) xnk is the transmitted data at the th k subcarrier of the th n OFDM symbol; (, ) H nk is the channel frequency response coefficient, and (, ) wnk is AWGN which having zero mean and unite variance. To perform decision-directed channel estimation, the latest channel information from the ( 1) th n  OFDM system is used to equalized the received signal by ' ˆ ˆ( , ) ( , )/ ( 1, ) x nk ynk H n k   . (2) Let ˆ( , ) xnk denote the from ' ˆ( , ) x nk . The conventional method [2] then uses ˆ( , ) xnk to estimate the channel frequency response by ˆ ˆ ( , ) ( , )/ ( , ) H nk ynk xnk  (3)   ˆ ˆ ( , ) ( , ) / ( , ) ( , )/ ( , ). xnk H nk xnk wnk xnk    Examining (3) we can find that the estimated subcarrier channel response coefficient has noise component which is dependent on the constellation of ( , ) xnk . If ( , ) xnk is generated from inner constellation of a high-order modulation scheme, it may amplify the noise effect, causing significant error in (3). In order to improve the performance of decision- directed channel estimation in OFDM-based DVB-T2 systems, we propose to apply the following weighted least-square(WLS) method to estimate the channel impulse response vector h :  On the Improvement of Decision-directed Channel Estimation in DVB-T2 Systems Chen-Ting Chen, Min-Kuan C. Chang and Kuo-Guan Wu Graduate Institute of Communication Engineering, National Chung Hsing University, Taiwan 978-1-4673-6199-6/13/$31.00 ©2013 IEEE 2013 IEEE 17th International Symposium on Consumer Electronics (ISCE) 151