26-04-2012, 11:08 PM
O.K. a mini-tutorial on filters. I have tried to keep this brief: there is a lot more to filters than what follows . . . .
All filters are designed to have a characteristic impedance. The word 'impedance' here is a bit misleading, since the word 'resistance' is more accurate. There are a number of characteristic response curves that can be obtained for any filter requirement, depending on the 'base model' that is used: the type of characteristic required determines that choice. Amongst these choices are the three main classic types: Butterworth, Chebychev and Bessel. Amongst the others, a Cluer response is quite popular these days. The Butterworth gives a reasonably flat pass-band response, and its attenuation rolls off at a fairly moderate rate: typically 10 dB / decade. It's popular for non-critical applications and is often the easiest to design. A Bessel has a uniform phase response, but has a really slow attenuation rate: almost only ever used for filtering pulse-type waveforms. The Chebychev has the fastest attenuation rate of those three main types, but has a ripple in the pass-band. To design any filter, certain parameters must be known: pass-band and stop-band freqs., the resistance that the filter operates between, the atten. in the pass-band, the amount of atten. required at a given freq. outside of the stop-band - and in the case of the Chebychev, the amount of ripple that is acceptable. Many filters do give a response outside of the pass-band - and this is usually a level that is significantly less than that in the pass-band, and sometimes at more than one freq. - the Chebychev and especially the Cluer being prime examples - and sometimes this response level can be 'input' as a chosen parameter. And that's where the trouble / skill kicks in, since many of those 'input' choices are not independent. And even after the maths has been done, it's possible to end up with values of L and C that are not realisable in practice: things like 0.1 pF or 100 Henries! In which case, it's back to the drawing board.
As you say, there are several filter design utilities out there on the 'Net. If you play around with them, aiming to design a filter, you'll get a better appreciation of what I've written above. (I did my filter design 'apprenticeship' long before PCs were in such an abundance!
)
As far as using a transformer to achieve a match, it really depends on its impedance characteristics - and especially the uniform (or otherwise) behaviour of those characteristics - over the band of freqs. of interest: pass-band and stop-band. That's why a purely resistive source / load is often chosen: makes the sums easier; eliminates a few more unknowns.
An example of using such a transformer is, of course, in an I.F. amplifier. The 'seen' impedance of the L/C ccts. will be high, and in a valve cct., the source and load impedances will be high also - and usually resistive. The fact that the pass-band (in this case a BPF is required) is relatively narrow compared to the centre freq., makes the design a lot easier. However, when we come to transistor I.F. amplifiers, the O/P Z of the driving transistor is often too low to effect a good match. Hence, the collector is often tapped down the primary winding in order to prevent the characteristic Z of the filter from being upset too much. In both cases, (valve and transistor) the fact that there is a mutual coupling, pri. / sec. does not alter the requirement for matching: it just enables a BPF response of the required shape to be obtained: the driving source simply 'sees' a load that is frequency dependent with a particular operating impedance and a particular pass-band / stop-band combination.
HTH
Al.
All filters are designed to have a characteristic impedance. The word 'impedance' here is a bit misleading, since the word 'resistance' is more accurate. There are a number of characteristic response curves that can be obtained for any filter requirement, depending on the 'base model' that is used: the type of characteristic required determines that choice. Amongst these choices are the three main classic types: Butterworth, Chebychev and Bessel. Amongst the others, a Cluer response is quite popular these days. The Butterworth gives a reasonably flat pass-band response, and its attenuation rolls off at a fairly moderate rate: typically 10 dB / decade. It's popular for non-critical applications and is often the easiest to design. A Bessel has a uniform phase response, but has a really slow attenuation rate: almost only ever used for filtering pulse-type waveforms. The Chebychev has the fastest attenuation rate of those three main types, but has a ripple in the pass-band. To design any filter, certain parameters must be known: pass-band and stop-band freqs., the resistance that the filter operates between, the atten. in the pass-band, the amount of atten. required at a given freq. outside of the stop-band - and in the case of the Chebychev, the amount of ripple that is acceptable. Many filters do give a response outside of the pass-band - and this is usually a level that is significantly less than that in the pass-band, and sometimes at more than one freq. - the Chebychev and especially the Cluer being prime examples - and sometimes this response level can be 'input' as a chosen parameter. And that's where the trouble / skill kicks in, since many of those 'input' choices are not independent. And even after the maths has been done, it's possible to end up with values of L and C that are not realisable in practice: things like 0.1 pF or 100 Henries! In which case, it's back to the drawing board.
As you say, there are several filter design utilities out there on the 'Net. If you play around with them, aiming to design a filter, you'll get a better appreciation of what I've written above. (I did my filter design 'apprenticeship' long before PCs were in such an abundance!
)As far as using a transformer to achieve a match, it really depends on its impedance characteristics - and especially the uniform (or otherwise) behaviour of those characteristics - over the band of freqs. of interest: pass-band and stop-band. That's why a purely resistive source / load is often chosen: makes the sums easier; eliminates a few more unknowns.
An example of using such a transformer is, of course, in an I.F. amplifier. The 'seen' impedance of the L/C ccts. will be high, and in a valve cct., the source and load impedances will be high also - and usually resistive. The fact that the pass-band (in this case a BPF is required) is relatively narrow compared to the centre freq., makes the design a lot easier. However, when we come to transistor I.F. amplifiers, the O/P Z of the driving transistor is often too low to effect a good match. Hence, the collector is often tapped down the primary winding in order to prevent the characteristic Z of the filter from being upset too much. In both cases, (valve and transistor) the fact that there is a mutual coupling, pri. / sec. does not alter the requirement for matching: it just enables a BPF response of the required shape to be obtained: the driving source simply 'sees' a load that is frequency dependent with a particular operating impedance and a particular pass-band / stop-band combination.
HTH
Al.
