Mobil haberleşme sistemlerinde konuşma kodlama

Başlık çevirisi mevcut değil.

Tez No: 55623
Yazar: TÜRKER BİRSEN
Danışmanlar: PROF.DR. GÜNSEL DURUSOY
Tez Türü: Yüksek Lisans
Konular: Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
Anahtar Kelimeler: Hareketli iletişim, Ses kodlama, İletişim sistemleri, Mobile communication, Voice coding, Communication systems
Yıl: 1996
Dil: Türkçe
Üniversite: İstanbul Teknik Üniversitesi
Enstitü: Fen Bilimleri Enstitüsü
Ana Bilim Dalı: Belirtilmemiş.
Bilim Dalı: Belirtilmemiş.
Sayfa Sayısı: Belirtilmemiş.

Özet

ÖZET Konuşma kodlamada temel amaç, mümkün olan en az kanal kapasitesinde (minimum band genişliği), yüksek kalitede ve ucuz olarak konuşma iletimini sağlamaktır. Bu, mobil haberleşme sistemlerinde kullanılan konuşma kodlayıcılar için de geçerlidir. Ancak mobil sistemlerde kanal kapasitesi daha önemli olduğundan kodlayıcının daha da düşük bit hızlarında kodlama yapması istenir. Ayrıca kodlama karmaşıklığı da önemli bir sistem parametresidir. Bu çalışmada mobil haberleşme sistemlerinde kullanılan konuşma kodlama teknikleri anlatılmıştır. Bu kodlayıcılar mobil sistemlerde kullanılacağından gerekli şartlan sağlamalıdır. Bu şartlar düşük bit hızı ve düşük karmaşıklık gereksinimidir. Bunun için geliştirilen tekniklerin temelinde öngörüsel kodlama yatar. Parametre kodlama tekniği de yeni ve mobil sistemlere oldukça uygun bir tekniktir. Bu teknikler ile ilgili ayrıntılı bilgi, çalışmanın ilgili bölümlerinde verilmiştir. Bu çalışmanın Giriş“ bölümünde, mobil sistemler hakkında çok kısa bir bilgi verilmiştir. Daha sonra ikinci bölümde, kodlayıcıların sınıflandırılması yapılmıştır. Üçüncü bölümde, dalga şekli kodlayıcılar anlatılmıştır. Adaptif öngörü kodlayıcısı (APC) ve altband kodlayıcısı (SBC) dalga şekli kodlayıcılardır. Dördüncü bölümde, uyarmalı kodlayıcılar anlatılmaktadır Uyarmalı kodlayıcılara örnek ”Düzenli darbe uyarmalı kodlayıcı (RPE)“ ve ”Kod uyarmalı doğrusal öngörü kodlayıcısı (CELP)" dir. Bu kodlayıcıların bit hızlan oldukça düşüktür. Beşinci bölümde ise konuşma kodlayıcılarının performansını arttıran çıkış filtresi (postfilter) ve hata kontrol kodlaması anlatılmıştır.

Özet (Çeviri)

SUMMARY SPEECH CODING FOR MOBILE TELECOMMUNICATION SYSTEMS Mobile telecommunication is one of the favorite subject of the last decade. Many researches are being made and will be made in the future. Speech coding for mobile telecommunication systems is an important subject for research that is developing parallelly to mobile telecommunication. In this study different speech coding techniques for mobile telecommunication systems are mentioned. The basic aim in speech coding is to design the coder which performs at minimum channel capacity, maximum quality, and minimum cost. The cost of encoding depends on the complexity of coder, and complexity depends on the channel efficiency. Thus, we must design optimum coder according to application area. These requirements are also in common for speech coders in mobile telecommunication systems. Additionally, mobile telecommunication faces several problems that have not been addressed in other telecommunication systems. The speech coder which is used in this application area must satisfy a variety of extra requirements. The most prominent of this is low bit-rate. Another important parameter is encoding delay. We can classify the speech coders into two fundamental groups. The first is, waveform coders, and the second is excitation coders. Waveform coders can be also classified into two subgroup as time domain coders and frequency domain coders. PCM, ADPCM, APC (Adaptive Predictive Coder), and SBC (Subband Coder) are included to waveform coders. Excitation coders, which are reffered as analysis- synthesis or parameter coders, are RPE (Regular Pulse Excitation), MPE(Multipuls Excitation), and CELP(Code Excited Linear Predictive) coders. APC, SBC, RPE, and CELP coders are mentioned in this study. Traditionally, toll quality coders have been waveform coders. Among waveform coders are the log-PCM coders, and the more complex ADPCM coding techniques. ADPCM schemes employ an adaptive predictor and an adaptive quantizer matched to the short-term modes of stationarity of the input speech. These strategies therefore exhibit improved performance over PCM schemes at the expense of greater complexity. The ITU-T has formally approved an ADPCM coding algorithm that provides toll quality speech at 32 kbit/s. This encoding algorithm utilizes backward adaptation schemes or the predictor and quantizer, i.e., the adaptations are based on an analysis of the past quantized data. This data is also available at the decoder, and therefore the receiver does not require side information. VIThe use of backward adaptation allows for very low encoding delay. In contrast, coders utilizing forward adaptation, the predictor and quantizer parameters are determined from an analysis coder input data. Since the analysis usually uses a 16-20 ms block of speech data, the input data has to be buffered This buffering results in a large encoding delay The block diagram of a generalized predictive coder with a short-term or formant predictor and a long-term or pitch predictor is shown in Fig. I. This configuration allows for adaptive adjustment of the noise spectrum in relation to the speech spectrum. The shape of the noise spectrum in relation to the speech spectrum is important from the point of view of perceived distortion in the output speech. Noise in the formant regions is partially or totally masked by the speech signal, since the speech power is high in the formant regions. The perceived noise in the output speech therefore comes from noise in those frequency ranges where the signal level is low. In Fig. 1 F(z) and N(z) are given by F(z) = £aiz-i, N(z) = İbîz-1 (1) i=I i=l For the configuration shown in the Fig. 1 the spectrum of the reconstruction error is given by sw-*w-cw-j^M (2) where the quantization error spectrum is given by Q(z). With the usual assumptions of uncorrelated quantization noise, the quantization error has a flat spectrum can be controlled by choosing N(z) appropriately. It is usually to choose N(z) as bandwidth expanded version of F(z). i.e., N(z)=F(z/ui), where 0<U<1. A value of u;=l gives a white reconstruction error spectrum, while \i=0 gives an error spectrum which has the same shape as the signal spectrum. The value of u: is usually chosen to be between 0.75 and 0.9. Such a value of u. has the effect of decreasing the noise power in the valleys of the speech spectral envelop, and increasing the noise power in the formant regions. The use of pitch prediction is motivated by the fact that voice speech segments exhibit considerable similarity between adjacent pitch periods. Transfer function of a 3 -tap pitch predictor is given by P(z) = P,z"Mp+1 +p2z-M' +p,3z~Mp~1 (3) where Mp is the estimated pitch period in samples. Adaptation of pitch predictors is necessary, since both the pitch lag and the predictor coefficients have to be fine tuned to analysis segment. VIISince the signal spectrum is time varying, the coefficients of the filter F(z) must be time varying also to obtain a small prediction error variance. The update algorithm used for the formant predictor can be an adaptive lattice algorithm..SPEECH s(n) 1-F(Z) N(Z) m® q(n) ?& e(n) QUANTIZER ea(n) DIGITAL -* LINK ^ P(Z) (a) DIGITAL LINK *© eq(n) £ P(Z) <> Rz) (b) Fig. 1 Generalized predictive coder with pitch and formant prediction (a) Encoder (b) Decoder The pitch predictor adaptation scheme is a nonrecursive procedure operating on the past quantized formant residual samples. Adaptation of the pitch predictor requires both the pitch lag and the pitch coefficients to be updated. Conventionally, pitch prediction uses forward adaptation but, backward adaptation can be used to decrease encoding delay. V11IAnother robust technique in coding the speech for mobile telecommunication is subband coding (SBC). Subband coders can yield very good speech quality while offering a significant complexity advantage at low bit rate (e.g. 12kbit/s). The block diagram of an SBC system is shown in Fig.2. The incoming speech signal, sampled at an 8 kHz sampling rate, is divided into 16 separate frequency bands by passing the speech signal through a bank of parallel filters. The purpose of separating the speech signal into bands is to allocate the bits used for coding the speech signal among the bands to effect the best speech quality. This partitioning of the bits among the bands is done dynamically to fallow the changing characteristics of the speech waveform. The bit allocations are done based on the relative distribution of signal energies among the bands; bands with more energy are allocated more bits to accommodate the greater dynamic range for those signals. The outputs of the filters are band-limited to approximately 1/16 th. of the full signal bandwidth. And the sampling rate for this signals is reduced by a factor of 16. These decimated signals are then passed to the coder for that band. The number of levels in the quantizer is determined by the number of bits allocated to coding of the samples for that band. The quantizer step sizes are adjusted by the band energy values to match the range of sample values. The energy values are coded and transmitted with the codes for the sample values for each band. The receiver decodes the energy values and based on these band energies determines the bit allocation for the sample codes among the bands. The data stream corresponding to sample values can then be demultiplexed and sent to the appropriate band decoder. Each band decoder output is adjusted by the corresponding band energy value. The band samples are returned to the original sampling rate by an interpolation process that insert 15 zero valued samples between each of the decoder sample output values and passes them through the corresponding bandpass filter. The zero inserting causes the spectrum of the band-limited signal to be replicated across the full 4 kHz bandwidth and the filter removes all but the one replication that occurs at the correct frequency. The outputs of the 16 filters are finally summed to produce the reconstructed speech signal at the original 8 kHz sampling rate. Subband coding is able to achieve high speech quality because it adjusts the coding of signal on a band-by-band basis rather than designing one coder to accommodate the signal in all bands. The coding of the energies plays an important role in the speech coder because the spectral shape is principally responsible for the intelligibility of the speech signal Realistically, it is undesirable to use any more than 3000 bit/s to code the subband side information in a medium rate speech coding implementation (9.6-16 kbit/s) to allow for adequate bit resources to code the subband samples. In order to design a digital speech system that is robust to channel impairments, it is necessary to thoroughly investigate the effect of channel errors on the subband side information. IXSpeech t(p\ 8 kHz Sampling ?+BPF.1 BPF.2 Channel BPF.n Encoder Encoder Encoder Mux Energy Calculati on and bit allocation (a) Demux Decoder r* * Decoder Decoder EnerQi Decode and Bit Allocation 00 BPF.Î BPF.2 BPF.n Channel Speech Fig. 2 Subband coding of speechThe excitation coders which encode the parameters of the speech signals differ from the waveform coders. The RPE and CELP coders which are included to excitation coders are mentioned in this work. The basic coder structure of the RPE can be viewed as a residual modeling process, as shown in Fig. 3. In this figure, the residual r(n) is obtained by filtering the speech signal s(n) through a pth-order time varying filter A(z), A(z) = l + j>kz-\ (4) which can be determined with the use of linear prediction (LP) techniques. The difference between the LP-residual r(n) and a certain model residual v(n) is fed through the shaping filter l/A(z/y), 1 1 A(Z/Y) l + İAT1*-4 k=l 0<y<1 (5) This filter, which serves as an error weighting function, plays the same role as the feedback filter in adaptive predictive coding with noise shaping and the weighting filter in multipulse excitation (MPE) coders. The resulting weighted difference e(n) is squared and accumulated, and is used as a measure for determining the effectiveness of the presumed model v(n) of the residual r(n). The same excitation sequence v(n) is generated at the receiver and, the output speech is produced by filtering this sequence. The filter coefficients are encoded and transmitted through the channel. This speech coding technique,which is used in GSM, decrease the coding rate significantly. The other excitation coder is Code Excited Linear Predictive (CELP) coder. This is a powerful technique for the realization of high quality speech coders operating at bit rates in the range from 4 to 16 kbit/s. Much research has been done on reducing the complexity of this technique without comprising speech quality. A block diagram of the CELP coder structure is shown in Fig. 4. The CELP coder is one of the analysis - synthesis coders. The reconstructed speech is synthesized by filtering the excitation signal through a synthesis filter based on short- term prediction coefficients. The parameters of this filter are determined from the input speech signal, and are updated at regular intervals. The excitation signal is formed by the weighted sum of contributions from codebook. The codebook indexes and the weighted coefficients are determined by means of an analysis- synthesis procedure. For all possible combinations, the corresponding output signal is computed and compared to the input signal. The optimal set of codebook parameters is defined to be the set that minimizes the mean squared error of this difference signal. XIS(n) LP Analysis S A(z) Excitation Generator m r(n) + 1 A(z/y) k.b 00 Error Minimization <n) Excitation Generator (a) v(n) 1 AT» S(n) (b) Fig. 3 Block diagram of the regular pulse excitation coder (a) Encoder (b) Decoder Some useful techniques can be used for improving the speech quality. Channel coding is a major technique that protects the data from the channel impairments. In the case of high sensitivity to the channel noise,as in SBC, the channel coding gives high performance. In SBC, side information must be strongly protected from the noise, otherwise, the output speech cannot be reconstructed even if 1-bit error occurred. Quadrature Mirror Filter (QMF) and Postfilter are powerful techniques that are used in speech coding. These techniques improves the speech quality without comprising the bit rate. XllSpeech h k - G, (a) 0) Fig. 4 Block diagram of a CELP coder (a) Encoder (b) Decoder Xlll

Benzer Tezler

Tez No
78220
Gezgin haberleşme sistemlerinde konuşma kodlama
Speech coding for mobile communications systems
HALİME USTA
Yüksek Lisans
Türkçe
1998
Elektrik ve Elektronik Mühendisliği Kocaeli Üniversitesi
Elektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
PROF. DR. HASAN DİNÇER
Tez No
101198
GSM şebekesinde radyo olanaklarının yönetim ve kontrol protokolları
Radio resource management and control protocols in GSM network
BUKET SUCU
Yüksek Lisans
Türkçe
2000
Elektrik ve Elektronik Mühendisliği İstanbul Teknik Üniversitesi
PROF.DR. GÜNSEL DURUSOY
Tez No
83099
GSM sistemlerinde uzaktan kontrol
Remote control on GSM system
BÜLENT ÖNEN
Yüksek Lisans
Türkçe
1999
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol İstanbul Teknik Üniversitesi
Bilgisayar Mühendisliği Ana Bilim Dalı
DOÇ. DR. MEHMET BÜLENT ÖRENCİK
Tez No
46282
Hücresel mobil haberleşme sistemlerinde tasarım yöntemleri
Başlık çevirisi yok
LEVENT GERDAN
Yüksek Lisans
Türkçe
1995
Elektrik ve Elektronik Mühendisliği İstanbul Teknik Üniversitesi
PROF.DR. GÜNSEL DURUSOY
Tez No
104153
2.ve 3. nesil CDMA temelli mobil haberleşme protokolleri
2.and 3. generation mobile communication protocols based on CDMA
BARIŞ YAVUZ
Yüksek Lisans
Türkçe
2001
Elektrik ve Elektronik Mühendisliği İstanbul Teknik Üniversitesi
PROF. DR. GÜNSEL DURUSOY

Geri Dön