Winbond - Voice CODEC FAQ's

Voice CODEC FAQ's

· What does “CODEC” mean?
· What is 0TLP? What is a “Digital Milliwatt?” Are there standard audio input and output levels?
· What is dBrnC0 and what is ICN?
· What does C-Message and Psophometric weighting mean?
· What are the various G numbers mean, such as G.712, G.711, etc.?
· What is the difference between µ-Law and A-Law? What are the benefits of these coding schemes?
· What is Gain Tracking?
· Why do the filters have such tight specifications?
· What is Frame Sync?
· What is the “Bit Clock?”
· What difference does a higher bit clock rate make?
· What does the CODEC connect to on the digital side? Why do I need a pull-up resistor on DOUT?
· During what state of BCLK is data in and data out valid?
· What is “sidetone?”
· Is echo cancellation required for a CODEC?

--------------------------------------------------------------------------------

· What does “CODEC” mean?

CODEC is a contraction of COder and DECoder. The word describes the two basic functions of digital communications. An (A)nalog signal is coded (converted) into a (D)igital equivalent in the A to D converter. At the other end of the digital circuit, it is decoded from the (D)igital equivalent, through the D to A converter, back into an (A)nalog signal. [There are many kinds of CODECs for voice, video and other analog signals but this page is about voice telephony CODECs. In this set of FAQs the word CODEC implies a single channel function. CODECs come in single, dual and quadruple packages but here we will consider the CODEC as a single channel function.]

--------------------------------------------------------------------------------

· What is 0TLP? What is a “Digital Milliwatt?” Are there standard audio input and output levels?

In voice telephony applications (and professional audio, as well) a reference level is needed. This level is to establish and maintain quality throughout the system. 0TLP means Zero Transmission Level Point. (Also called 0dBm0) In most audio applications this level is also 0 dBm, or .775 VRMS (2.19 VPP) into 600 ohms impedance. This is one milliwatt of audio energy. Just as there is a reference level in the analog domain, there needs to be one in the digital domain. This digital reference for PCM telephony is called the Digital Milliwatt and is represented by a specific code. When this digital code is sent into a CODEC a 0TLP level 1020 Hz analog tone is output from the chip. This tone will be at 0TLP or 0dBm0.

--------------------------------------------------------------------------------

· What is dBrnC0 and what is ICN?

ICN is Idle Channel Noise, or the noise present in a circuit output with no input signal present. [Or when a PCM zero code is received.] It is usually measured through the C-, or Psophometric, weighting curves and referenced to either the 0TLP or to 90 dB below 0TLP (1 picowatt.) Using the North American technique, an expected value of ICN might be 11 dBrnC0. This is explained as being 11 dB greater than –90 dB reference noise when measured through a C-Message weighting filter referenced to 0TLP.

--------------------------------------------------------------------------------

· What does C-Message and Psophometric weighting mean?

These terms are the names of two similar, but slightly different, weighting curves used to determine the noise in a communications circuit. One is for North America and Japan (C-Message), and the other is for Europe and the rest of the world (Psophometric). This is just like the PCM encoding laws. Rather than measuring broadband noise through a communications circuit flat (That is, giving equal weight to all frequencies) the noise is passed through a filter that is shaped to reflect the human ear’s response to interfering frequencies. Thus, a noise at 50 or 60 Hz is given less “weight” than a noise at 1000 Hz. It is felt that this is a more realistic measurement of noise. The shapes of these curves are defined in ITU-T O.41 Annex A.

--------------------------------------------------------------------------------

· What are the various G numbers mean, such as G.712, G.711, etc.?

The G specifications are one part of a vast series of Recommendations by the Telecommunication Standardization Sector of the ITU (International Telecommunication Union) They have 26 chapters in their series; one for each letter of the alphabet. The G Series are Transmission Systems and Media, Digital Systems and Networks and it is this series that is most applicable to CODEC devices. G.711 and G.712 define the CODEC performance required.

--------------------------------------------------------------------------------

· What is the difference between µ-Law and A-Law? What are the benefits of these coding schemes?

µ-law and A-law are audio compression schemes defined by ITU-T G.711 that compress 16 bit linear data down to 8 bits of logarithmic data. The encoding process (referred to logarithmic companding) breaks the linear data into segments with each progressively higher segment doubling in size. This ensures that the lower amplitude signals (where most of the information in speech takes place) get the highest bit resolution while still allowing enough dynamic range to encode high amplitude signals. Though this method does not provide a very high compression ratio (roughly 2:1), it does not require much processing power to decode. Mu-law (also written µ-Law) is the encoding scheme used in North America and Japan for voice traffic. A-Law (or a-Law) is used in Europe and throughout the rest of the world. The two schemes are very similar. Both break the total dynamic range into eight positive and eight negative segments. Bit 1 (MSB) identifies the polarity, bits 2,3,4 identify the segment, and the last four bits quantize the value within the segment. The differences are in the actual coding levels and the bit inversion. Nevertheless, both systems offer 2:1 bit compression, thus doubling the capacity of a digital transmission circuit while maintaining “toll quality” voice reproduction.

--------------------------------------------------------------------------------

· What is Gain Tracking?

Gain Tracking is the measurement of the linearity of the Encoding and Decoding processes. Using a reference level of –10 dBm0, a 1020 Hz signal is varied in level from +3 to –55 dBm0 and expected to remain within a particular tolerance envelope. This tolerance envelope is defined in ITU-T G.712 and, as an example, is +/- 0.5 dB from +3 to –40 dBm0 and +/- 3 dB at –55 dBm0.

--------------------------------------------------------------------------------

· Why do the filters have such tight specifications?

The filters in CODECs have very tight specifications in order to maintain quality over multiple “hops” or links. Any error in a CODEC filter could be cumulative and over many hops could make a particular frequency fall out of specification.

--------------------------------------------------------------------------------

· What is Frame Sync?

Frame sync(hronization) is the 8 KHz pulse that is presented to the CODEC to maintain its sampling frequency and sequence. Typically, a frame sync pulse is presented to a CODEC when it is supposed to feed out its 8 data bits at the data (bit) clock rate onto the PCM highway. The CODEC occupying the next time slot will get its frame sync pulse 8 data bit clocks later so that the following 8 data bits come from that CODEC.

--------------------------------------------------------------------------------

· What is the “Bit Clock?”

The Bit Clock is the clock that determines the rate of the data bits in and out of the CODEC digital interface. This typically is 1.544 MHz in a North American system and 2.048 MHz in a European system. [See next question and answer.]

--------------------------------------------------------------------------------

· What difference does a higher bit clock rate make?

The higher the bit clock rate in a system the more voice channels can be carried on the PCM highway. In a single channel system the bit clock can be as slow as 64 KHz, which will accommodate 8 bits of data every 8 KHz (125 µseconds.) As the bit rate on the highway is increased, more voice channels can be carried. A 256 KHz bit clock will carry four voice channels and a 2.048 MHz bit clock will carry 32 voice channels.

--------------------------------------------------------------------------------

· What does the CODEC connect to on the digital side? Why do I need a pull-up resistor on DOUT?

The digital output of a CODEC usually connects to the digital highway within a switch, network trunk or DSP chip. Each CODEC carries one voice channel and as many as 24 (NA) or 30 (Europe) will be connected together (in parallel) to the digital highway. Each output becomes active (open drain) only during it’s time slot for 8 bit times and then goes tristate while the other CODECs output their 8-bit words in their assigned time slots. Therefore, a single pull-up resistor is required to enable the outputs to go HIGH to represent a ONE level.

--------------------------------------------------------------------------------

· During what state of BCLK is data in and data out valid?

Normally the PCM data input must be valid on the falling edge of the BCLK and the output is valid on the rising edge of the BCLK.

--------------------------------------------------------------------------------

· What is “sidetone?”

Sidetone is that amount of the voice sent that is heard in the earphone of the same telephone set. This enables the person to hear themself talk. From the New Zealand telecommunication specification; “Sidetone performance results from human head characteristics in combination with the mechanical and electro-acoustical properties of the telephone. Some sidetone is important to avoid "deadness" on a telephone, but excessive levels of sidetone have a marked effect not only on the perception of received signals, but also on the volume of the speaking person’s voice level. Thus, excessive sidetone in one telephone can have a marked effect on the received speech volume of the telephone at the other end of a call.”

--------------------------------------------------------------------------------

· Is echo cancellation required for a CODEC?

No, echo cancellation is not done by the CODEC. However, echo cancellation is usually done with a CODEC providing the digital input to a DSP processor that does the echo cancellation.