Mobile Radio Link Adaptation by Radio Channel State Prediction

The paper explains the prediction range adaptation and optimal setting of the sampling period of the radio signal envelope to guarantee optimal quality and capacity of the radio channel. The developed algorithms FSPRA and VSPRA provide optimal setting of the prediction range with respect to the mobile station motion. Both algorithms are compared with the algorithm without adaptation. Efficiency of algorithms and their contribution to minimization of the radio channel state prediction error, optimal prediction gain for a sufficiently large prediction range with respect to execution speed of the adaptation algorithm are evaluated. The paper focuses on using the Kalman filter to predict the radio signal fading envelope and potential implementation in the transmitter power control loop. Ill. 4, bibl. 10 (in English; abstracts in English and Lithuanian).


Introduction
Applying the channel state prediction in 3G mobile communication networks [1,2] allows securing a maximum spectral efficiency of the channel under the required error rate.
The paper deals with the selection analysis of the optimum sampling period of the radio channel signal envelope [3] and prediction interval adaptation using Kalman filter to achieve minimum channel state prediction error.Results of radio channel state prediction simulations performed in Matlab® are applicable in optimization of adaptation parameters of a real communication system [4].

Simulation of Kalman filter as radio channel state predictor
Kalman filter enables parameters prediction also for non-stationary signals [5,6] using all information about the measured signal from the start of measurement or signal reception.The predictor applies a filter with a recursive structure.Kalman filter belongs to adaptive predictors because its coefficients are adapted in each step so as to provide optimal estimate of the signal.The predictor performs two repeatedly changing steps -prediction (time update) and correction (measurement update).

Algorithm for radio channel state prediction using Kalman filtering
Using the Kalman filter method, the following steps are performed in each simulation step:  Based upon the specified sampling period of the radio signal envelope sampl where L denotes the prediction range, i.e. the number of future samples for which the prediction is defined;  The prediction error is evaluated using the formula and 24 kHz), for which the best quality of prediction, i.e. the lowest value of the prediction error was achieved [7] (Fig. 1).With regard to simultaneous utilization of predictors in the link adaptation loop it is important to know properties of the predictor with Kalman filter also for various prediction intervals, i.e. prediction range scale L .The received signal processing, generating of the related TPC link adaptation command and its transmission through the control channel takes certain time, hence the control station works with a relatively out-dated information about the radio channel state.
Elimination of the total delay total n is the main objective of implementation of prediction methods for radio channel state estimation into transmission power control algorithms.To simulate real system, for sampling frequency sampl f = 12 kHz, the UMTS time window period was chosen, i.e.In this way, limits for the simulation without adaptation have been obtained, that can be further compared with results from simulation with prediction range adaptation (FSPRA and VSPRA algorithms).
From simulations results that the smaller is the sampling period of the radio signal envelope, the better prediction is achieved [8].A drawback of a small radio signal envelope sampling period consists in hardware demands for computation unit and prediction possibility over a very small prediction interval; this lowers its applicability in the link adaptation loop.It is quite complicated to carry out real-time simulation using sampling frequency 24 kHz to guarantee the required QoS criterion in various types of communication environment and for various speeds of the mobile terminal.
Similar results can be achieved using methods for optimal tuning of the prediction range scale L and the sampling frequency sampl f (algorithms FSPRA and VSPRA with adaptation).L .In our case, the prediction range step corresponds to time between two subsequent samples of the radio signal envelope, which corresponds to the sampling period of the radio signal fading envelope.

Realization of the
To objectively compare simulation results, the mobile station moved in heterogeneous environment.Because the environment parameters depend on the time-varying mobile station velocity, the system has to periodically update the adaptation process with the period of the UMTS time window, i.e. every 0.667 ms.Simulation was performed for two sampling frequencies of the radio signal envelope sampl f = 12 kHz and 24 kHz.The required prediction error value was allow  = 1 dB.

Variable step prediction range adaptation
The Variable Step Prediction Range Adaptation algorithm (VSPRA) is the second method for optimal changing of the prediction range parameter The objective evaluation of simulations requires securing equal conditions, hence also in these simulations the mobile station moved in a heterogeneous environment.The adaptation process was periodically updated with the period of UMTS frame time window (0.667 ms).Simulation was carried out for two sampling frequencies of the radio signal envelope sampl

Evaluation of simulation results
Results of the radio channel state prediction simulation without prediction range adaptation, with fix step adaptation (FSPRA algorithm) and with variable step adaptation (VSPRA algorithm) are compared in Fig. 2 using cumulative distribution function CDF.
Comparison of FSPRA and VSPRA adaptation algorithms shows reduction of the prediction error error  when using the variable step prediction range adaptation    The paper explains the prediction range adaptation and optimal setting of the sampling period of the radio signal envelope to guarantee optimal quality and capacity of the radio channel.The developed algorithms FSPRA and VSPRA provide optimal setting of the prediction range with respect to the mobile station motion.Both algorithms are compared with the algorithm without adaptation.Efficiency of algorithms and their contribution to minimization of the radio channel state prediction error, optimal prediction gain for a sufficiently large prediction range with respect to execution speed of the adaptation algorithm are evaluated.The paper focuses on using the Kalman filter to predict the radio signal fading envelope and potential implementation in the transmitter power control loop.Ill.4, bibl.10 (in English; abstracts in English and Lithuanian).Pateikiamas optimalią kokybę ir radijo kanalo talpą garantuojančio radijo signalo prognozuojamas veikimo nuotolis ir optimalūs nustatymai reikiamu laikotarpiu.Analizuojami FSPRA ir VSPRA algoritmai, įvertinantys mobiliojo ryšio stoties padėtį, nusako optimalius prognozuojamo veikimo nuotolio parametrus.Pateikti algoritmai yra palyginti tarpusavyje bei su nepristaikančiuoju algoritmu.Pateiktas algoritmo efektyvumas, radijo kanalo būsenos prognozavimo klaidų mažinimo būdai ir kt.Daugiausia dėmesio skiriama radijo signalo būsenai prognozuoti Kalmano filtru.Il. 4, bibl.10 (anglų kalba; santraukos anglų ir lietuvių k.).

f
, a set of input samples   n Z is generated over the time T ;  Based on the input data   n Z , Kalman filter predicts the future state of the radio signal envelope   L n I  ~, e. calculating the difference between the predicted value and the real one of the radio signal envelope level;  Prediction error is evaluated after the time T (adaptation period) elapses, in our simulation T is the size of the UMTS time slot, i. e. 0.667 ms;  Optimal setting of the prediction range scale L , i.e. the defined time of prediction, and radio channel envelope sampling frequency sampl f to guarantee the required prediction error level allow  .To be able to objectively evaluate obtained simulation results, the first simulations were carried out without adaptation of parameters of the prediction range scale L , with sampling frequency of the radio signal envelope sampl f .The parameters scale L and sampl f were constant during simulation.Two sampling frequencies of the radio signal envelope were chosen ( sampl f = 12 kHz

Fig. 1 .
Fig. 1.Mean value, standard deviation and covariance of the prediction error versus radio channel envelope sampling period prediction range).The prediction range for sampling frequency sampl f = 24 kHz was chosen as the period between two subsequent samples of the radio signal envelope, i.e. 041 .0  scale L ms (minimum prediction error).
Fix Step Prediction Range Adaptation (FSPRA) algorithm consists in guaranteeing minimum value of the adaptation criterion error  by changing the prediction range parameter scale L .The FSPRA algorithm is based upon evaluation of the current value of prediction error error  and consecutive increasing/decreasing of the prediction range scale L by a fix adaptation step step scale L in order to minimize the adaptation criterion value error  .The main principle of the prediction range adaptation block with variable step is based upon evaluation of the current value of the prediction error error  for the given prediction range, as well.Two parameters were introduced, affecting the prediction range change scale L  based on the evaluation of the current value of the prediction error error of the prediction range scale L  , namely decrease, without change and of growth.Another qualitative parameter is the change weight versa, the value 1 indicates the highest possible slope of the prediction range change.The change weight value of the prediction error was again allow  = 1 dB.
Fig. 2. Compa without predic algorithms FSPR Predictio prediction err algorithms FS The largest pr VSPRA algor kHz.The pred period of the U From th concludes that s achieved (V kHz), work w when the mo According to t prediction rang the required pr Compari application in prediction gai n Fig. 4. The algo 24  sampl f kHz achieves the b G(L) and on channel envel Fig. algor Fig. algorCon