This preview shows pages 1–3. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: ECE 421 - Sum 2010 Notes Set 14: Linear Prediction Coding 1 INTRODUCTION Bandwidth is a very limited resource in wireless voice and data systems. Storage is also a limited resource, especially on portable and handheld devices. Additionally, short packets generally cause less congestion in networks than do long packets. For these reasons it is frequently desired to send an information signal using as few bits as possible. However, decreasing the sampling rate or the storage size frequently is not neces- sarily the solution for decreasing the bit rate. Simply sampling the analog signal slower will frequently lead to too much distortion in the reconstructed signal at the receiver. As an example, regular wireline Pulse Code Modulation (PCM) for telephone has a bit rate of 64,000 bps. With a modulation efficiency of 1 bps/Hz this would require a trans- mitted bandwidth of 64 KHz for one voice subscriber. This is quite a bit of spectrum, especially if numerous other subscribers must be supported at the same time. Thus, there is a great deal of interest in encoding the information source with as few bits as possible while retaining low distortion. This is called Source Coding. In Source Coding, we create encoding algorithms which are specific to the type of source. For example, the statistics and frequency domain properties of still pictures, motion pic- tures, and speech/audio are sufficiently different from each other that different source cod- ing algorithms have been developed for each of these information types: Still images (like photographs) may be encoded using JPEG algorithms, Sequences of images (motion pictures) may be encoded using MPEG algorithms, Speech may encoded using Linear Prediction Coding (LPC) algorithms In this set of Notes we will examine the some of the very basic properties of LPC for speech. LPC is sometimes called a lossy data compression algorithm, meaning there is some loss of fidelity compared to the original speech. LPC methods used in actual systems are more complicated than what we will examine, but we will study a sufficient amount of material to demonstrate some of the underlying principles. ECE 421 - Sum 2010 Notes Set 14: Linear Prediction Coding 2 SPEECH PROPERTIES LPC depends upon parts of speech (like vowel sounds) having some degree of periodicity or being predictable. For example, here is a plot of approximately 0.06 seconds of someone saying the number one: 100 200 300 400 500 600-1000-800-600-400-200 200 400 600 800 1000 The amplitude scale is the integer value produced by the A/D converter and is proportional to voltage. The horizontal scale is in samples, taken T s = 125 sec. apart. Note that there is a great deal of structure or periodicity in the waveform. It certainly does not look like a random noise waveform. A way of mathematically stating this property is to say this speech waveform has correlation....
View Full Document
- Summer '08
- Signal Processing