Yong Cheol Kim

Fingerprint Wi-Fi Radio Map Interpolated by Discontinuity Preserving Smoothing

In location fingerprinting, the position of a mobile device is identified by matching its received signal strength pattern with pre-recorded radio map. Interpolating a coarse radio map into a dense one may reduce the cost of constructing a radio map. An interpolated radio map, however, has low accuracy near a wall. We present a method of constructing a high-density radio map by discontinuity preserving smoothing (DPS). Adaptive smoothing preserves the discontinuity by localized smoothing in accordance with the wall layout of a building. Experimental results show that a DPS-reconstructed radio map has higher accuracy than other methods. With sampling ratio ≥ 20%, the performance is close to a genuine full density radio map. With sampling ratio > 60%, the performance is even better than the original full density map.

Reconstruction of radio map from sparse RSS data by discontinuity preserving smoothing

In Wi-Fi based location fingerprinting, the cost of constructing a radio map is high due to the calibration of measured RSS data at a large number of RPs and the update of database. Interpolating a coarse radio map into a dense one may reduce the cost. An interpolated radio map, however, has low accuracy, especially at space discontinuity such as a wall. We present a method of constructing a high-density radio map which preserves the discontinuity of RSS by localized smoothing in accordance with the layout of a building. Experimental results show that the radio map by discontinuity preserving smoothing has higher accuracy than conventional interpolating methods. With sampling density ≥ 35%, the performance is close to a genuine full density radio map. With sampling density > 60%, the performance is even better than the original full density map.

Hidden Markov Model Based Tracking of a Proxy RP in Wi-Fi Localization

In indoor localization of a mobile device (MD) in WLAN, a proxy reference point (RP) plays the role of representing the dynamically changing channel parameters between an access point (AP) and a MD. We present a hidden Markov model (HMM) based method of tracking the proxy RP for a MD. First, we examined the performance of finding the proxy RP for a stationary MD as we vary the number of APs, the deployment style of APs and the noise strength in RSS. Then, for the tracking of proxy RP for a moving MD, we developed a HMM-based algorithm, which provides the relation between the RSS vector and its proxy RP in a probabilistic manner. Both in computer simulation and in real measurements, HMM-based tracking proved to have much better performance than point-by-point decision by Euclidean distance between RSS vectors.

Processing Gain in a Recursive Single Parity Check Product Code with Non-Gaussian Weight Distribution

We present a recursive single parity check product code, which was originally introduced as a constant amplitude multi-code CDMA. The proposed code has two advantageous features over conventional parity check product codes. First, the proposed code does not have any low-weight codewords. We derived a closed-form solution for the weight distribution. The code weights are symmetrically distributed at (N plusmn radic(N))/2, where N is the full-length of a codeword. Second, the processing gain in the despreading process after the iterative decoding provides additional improvement in BER performance. As a result, the proposed scheme has excellent error correcting capability. When soft-output iterative decoding is applied, the performance is away from the Shannon capacity limit by only 0.95 dB. Hence, the proposed code is a simple coding scheme with a good error correction performance.

Differentiated assignment of extrinsic information based on code weights of single parity check product code

Constant amplitude multi-code CDMA (CAMC) removes the amplitude fluctuation in multi-code CDMA, by which the large power consumption of a power amplifier with a strict linearity can be avoided. The nature of CAMC is a recursive single parity check product code (SPCPC). CAMC for N = 4k is equivalent to k-dim SPCPC. As a top-level CAMC codeword is recursively constructed from lower-level codewords, log likelihood ratio (LLR), a priori information (API) and extrinsic information (El) of upper-level codewords are obtained as a function lower-level codewords. The codewords of CAMC were found to have fixed code weights, (N ± √ N)/2. In this paper, we show that differentiated assignment of El in the iterated decoding can boost the BER performance. If any of the lower-level codewords have weights other than the fixed value, the El associated with these erroneous lower-level codewords has lower confidence. El from a wrong codeword is given lower weights in the computation of LLR, API and El of the upper-level codeword. With differentiated assignment of El, the BER performance improved by 0.1~0.3 dB.

Extrinsic information in iterated decoding of fixed weight codewords

Constant amplitude multi-code (CAMC) CDMA is generated by spreading of lower dimensional codewords concatenated with bit-by-bit parity vectors. CAMC has dual aspects of a spread spectrum signal and a product code. Error correcting methods for CAMC are based either on processing gain in spread spectrum, or on iterative decoding of product codes. The weights of CAMC codewords are fixed at two symmetrical values. In this paper, we present performance improvement of CAMC by the control of extrinsic information (EI) in the iterated decoding of CAMC. The control is based on whether the decomposed codewords have the correct weights, or not. Preferred assignment of EI on correct weight and as a predictor of despreading brings an improvement in BER performance of 0.1 ~ 0.3 dB.

Differentiated assignment of extrinsic information in iterated decoding of fixed weight codewords

Constant amplitude multi-code (CAMC) CDMA has the same structure as a recursively generated single parity check product code. A top-level codeword of CAMC is recursively constructed from lower-level codewords. In the iterative decoding of CAMC, log likelihood ratio (LLR), a priori information and extrinsic information (EI) of a codeword is a weighted sum of LLR values of associated codewords from which it is despread or into which it is spread. In this paper, we show that differentiated assignment of EI in the computation of LLR can improve the performance of bit error correction. The weights of CAMC codewords are fixed at two fixed values. We let EI converge fast to saturation value when a codeword has the correct weight. The proposed method achieved performance improvement of 0.1 ~ 0.3 dB in Eb/No over the regular iterated decoding of CAMC. When compared with despreading ON/OFF control, a gain of about 0. 1 dB is achieved, which is meaningful near the Shannon capacity limit.

Building Recognition Guided by Wi-Fi Based Localization

We propose a method of building recognition on imagery from a mobile device, when an approximate location of the user is available. The query image taken by a mobile device is matched with images of nearby buildings in database. We use SIFT descriptors for feature matching to get an invariance of scale and rotation between images taken at various shooting conditions. A rough localization of the mobile user is obtained by triangulation or fingerprinting of the received Wi-Fi signal strength from Wi-Fi access points. The proposed method is demonstrated on buildings with access points in vicinity. In the experiments, simulated street view images are generated to serve the role of on-line street view data. Results from our work on Wi-Fi based localization and radio map interpolation are also presented

Reasonable Resolution of Fingerprint Wi-Fi Radio Map for Dense Map Interpolation

In Wi-Fi based location fingerprinting, the cost of building a radio map is high. Though interpolating a coarse radio map into a dense one may reduce the cost, low accuracy at space discontinuity such as walls deteriorates the reliability of interpolated radio map. We present an interpolation of radio map, which preserves the discontinuity of received signal strength in accordance with the layout of a building. Furthermore, a closed form solution of reasonable resolution of radio map is derived. Experimental results show that, with sampling density >35 %, the localization accuracy is comparable to that of a full density map.

High Resolution Image Reconstructed by ARPS Motion Estimation and POCS

In POCS (projection onto convex sets)-based reconstruction of HR (high resolution) image, the quality of reconstructed image is gradually improved through iterative motion estimation and image restoration. The amount of computation, however, increases because of the repeated inter-frame motion estimation. In this paper, an HR reconstruction algorithm is proposed where modified ARPS (adaptive rood pattern search) and POCS are simultaneously performed. In the modified ARPS, the motion estimates obtained from phase correlation or from the previous steps in POCS reconstruction are utilized as the initial reference in the motion estimation. Moreover, estimated motion is regularized with reference to the neighboring blocks’ motion to enhance the reliability. Computer simulation results show that, when compared to conventional methods which are composed of full search block matching and POCS restoration, the proposed method is about 30 times faster and yet produces HR images of almost equal or better quality.