In earlier experimental systems this DC pedestal corresponded directly to an equivalent pedestal in the coder. General practice now is still to employ a pedestal or DC bias to enable unidirectional gates to be used, but otherwise to employ AC coupling and derive the centre point value by relatively long term integration.
In the demodulation or recovery stage the handling of narrow samples is a rather more involved problem than that of initial sampling. Theory declares that the recovered pulses are passed through a low pass filter. However , if sufficient power is to be delivered to avoid the need for considerable subsequent amplification of the analogue signal it is desirable, in order to avoid very high instantaneous power levels, to stretch the duration of the pulse. This results in a sinx/x reduction in the received level where x is the length of the stretched pulse expressed in radians as a fraction of one cycle f half the analogue frequency. Thus, if each pulse is stretched to the start of the subsequent pulse then at the Nyquist limit x=π/2. This degree of pulse stretching produces a theoretical 4dB loss at the Nyquist limit. This higher frequency attenuation characteristic can be corrected by the filter characteristic if extreme linearity is not of great importance.
There are many alternatives available . One popular procedure is to deliver the output pulse into a capacitor so that a particular voltage pulse transfers a corrected by the filter characteristic if extreme linearity is not of great importance.
There are many alternatives available. One procedure is to deliver the output pulse into a capacitor so that a particular voltage pulse transfers a corresponding amount of energy and this can be made high enough to limit subsequent amplification to a reasonable value. This is a satisfactory method but it requires careful engineering, since the reservoir capacitor is now in effect part of the filter and the frequency relationships must be appropriately chosen to ensure substantially complete transfer of the energy of each pulse.
(2)Quantising
The next problem to be examined is that of the distortion arising from the fundamental requirement of PCM that samples transmitted cannot be continuously variable but must be chosen from a finite set , the number of which is a function of the length of the binary number we are prepared to assign for the transmission of each sample.
Fig.3.5 shows a section of wave form being sampled. The permitted sampling
部分章节译文
approximations are indicated by the horizontal lines. It is assumed that the value selected is the nearest one + or — (in practice for reasons of convenience, the next lowest one is taken and at the recovery stage half a step value is added, which has identically the same result. ) The quantising errors are shown below . It will be seen that the mean value is ±1/4S where S is the magnitude of one step . Except where there is a fortuitous correlation between sampling rate and transmitted frequency , and this occurs infrequently in speech , the mean power is represented by:
(xdx)/S
1
S21 S2212S 12
Fig.3.5 Sampling of a wave form
The random nature of this distortion results in a frequency content which is substantially flat over the pass band of the channel.
This is generally called “quatising noise '”but since it is dependent on the existence of a signal it is a form of distortion as distinct from a constant background noise.
There does exist a form of noise (generally referred to as 'idle channel noise') related to quantising noise which is independent of the signal . If there exists an absolutely exact and stable match between the idle channel voltage and the reference voltage at the centre of the quantizing ladder then it would be possible to locate the idle channel condition exactly between two decision levels and until the peak-to-peak noise exceeds one quantum step it will not be quantized. As soon as it does, however, noise will be transmitted at a level corresponding to S. Since the circuit is dealing with a step of the order of 1mV, it is difficult to hold the pedestal relationship exactly where it should be and any deviation will tend to create an output of 212S even when the input 4
noise is less .
A manifestation of this very likely to arise in practice is the difficulty of guaranteeing that there will be no residual mains hum at this level . If such hum causes the apparent pedestal to move to and fro over a decision level apart from the direct quantisation of the hum , at a frequency of the 100 cycle order periods of extreme sensitivity to higher frequency noise will occur . There is thus a tendency to create a low level high frequency noise chopped at the hum frequency .
部分章节译文
This is subjectively more intrusive than steady noise . However , the self-centering arrangements and minor variations in gate behavior can cause fluctuations and resultant chopping at even lower and more noticeable frequencies . Optimal design is therefore a matter for rather careful compromise . In any case in systems now in use and planned , the idle channel noise at a level between -60 dB and -70dB is quite unnoticeable except in extremely quiet listening situations .