电子信息与通信专业外语(4)

(3) Companding

One of the major problems of practical PCM is the very wide range of power levels to be handled . The range within the speech of one speaker is some 25dB. Due to the differences between loud and quiet talkers and the attenuation of established connections before a PCM link is encountered , we are faced with a need to handle a range of some 60dB . If the lowest level to be handled just exceeds one step and peak clipping is reasonably limited then this 60dB on a uniform step basis would involve about +1000 levels .

This would not be catastrophic in regard to digital transmission rate (11 digits instead of the currently accepted eight digits). It would , however , reduce the economic attractions as far as pure transmission is concerned and it would greatly increase the cost and complexity of the terminal conversion processes .

To counter this difficulty we may take advantage of a practice which has become well established in many areas of conventional transmission, namely companding . Ininstantaneous companding the voltage of the wave form being transmitted is compressed in accordancewith an appropriate law and at the receiving end the inverse expansion occurs . Companding is used in the main in a somewhat complementary manner to lift the level on line of the low level signals while maintaining the peaks of the high level signal/noise ratio but at the expense of changing a constant level noise into one related to signal level in accordance with the companding law.

It must be recognized at the outset that voice signals are particularly amenable to this treatment because the probability distribution of voice energy levels tends to separate the high frequency low level fricatives from the relatively low frequency but high level vowels. Partly as a consequence of this and partly as a consequence of the subjective responses of the human ear and brain, speech communication is highly tolerant of noise levels directly related to the signal level . Much subjective testing by D.L. Richards and others has established that the vast majority of listeners are unable in the environment of telephony to detect signal dependent noise more than 25 dB below signal level, and even 20dB is noticed by relatively few.

To secure a linear relationship between quantising noise and signal level involves a logarithmic relationship between the number of steps and the signal level. A truly logarithmic curve would not pass through the origin (number of steps 0 and level 0) and there have been many curves examined which offer a reasonable compromise. Two have been the subject of much offer a reasonable compromise. Two have been the subject of much debate by the CCITT. These may be expressed as follows:

1 lgAx F(x) 1 lgA F(x) 1 lgAx

1 lgA

部分章节译文

law F(x) lg1( x) lg1( )

The A law is in effect truly logarithmic for higher levels with a linear bottom section comprising the tangent trough the origin.

The law has a comparable overall form but is nowhere truly logarithmic and nowhere truly linear, though it approximates to thee characteristics at the extremes.

It was established fairly early in studies that a seven- or eight-digit binary number giving a total of ±64 or ±128 levels was likely to be adequate and would not result in undue complexity or cost of implementation. The problem is one of compromise to hold in balance the following requirements:

① To handle high level voice signals with acceptable peak clipping. This means the onset of clipping should be in the region of +2 or +3 dB for a sinusoid at a point of zero equivalent.

② To maintain the signal/noise ratio over the upper levels at a figure in the 20 ~ 25 dB region or better .

③ To have the smallest quantum steps small enough to maintain transmission of low lever fricatives to the lowest practicable level. This means coverage of 60 dB or more.

④ To maintain idle channel noise in the region of 1000 W or less.

The use of the A law with A about 100 enables these requirements to be met with some margin assuming eight digits, i.e. ±128 levels.

The process of companding may be executed by a compression operation on the speech wave form followed by linear quantising and the inverse operation at the receiver. Suitable compressors can be constructed by employing the non-linear charcteristics of diodes and arranging a suitable diode/resistor network. This was done on many of the earlier systems. However , the specification and control of the diode characteristics is difficult , particularly in regard to the essential matching of complementary pairs . This becomes increasingly important when PCM switching enters the picture and any coder may be connected to any of a number of different decoders.

Later systems have therefore tended to use linear sampling and non-linear quantising and as a further aid to simple implementation to replace the smooth curve of the A or law by a series of straight line segments. With a reasonable number of segments the increase in quantising noise due to the departure from the smooth curve is negligible and this procedure does result in the cheapest implementation of high reproducible compression and expansion and the resultant satisfactory matching of any transmitter with any receiver.

百度搜索“77cn”或“免费范文网”即可找到本站免费阅读全部范文。收藏本站方便下次阅读，免费范文网，提供经典小说公务员考试电子信息与通信专业外语(4)在线全文阅读。