by Pete Goudreau
Preamps. We've all used them. We've all cursed them too, I bet. Little gets more complicated than the problems associated with connecting multiple sources to a control center and on to an amplifier or two. There's the ground-loop-induced hum, added noise and distortion from active gain stages, signal-level matching, tape-loop buffering, drive capability, the list goes on and on. Not an easy thing to achieve, perfection.
With the advent of digital sources, it seems just about every piece of equipment nowadays has a high-level output, generally always high enough to drive an amplifier to clipping when connected directly. So why would anyone want any additional gain in the preamp? Simple. Compression.
The ear responds to average loudness, not peaks. Compressing the signal raises the average level at the expense of dynamic range. Compression is extensively used by recording engineers to raise the average level of the recording to make it sound louder than all the others, and therein lies enhanced revenue. Therein also lies error
But what exactly is error? Even if the measurements clearly show a deviation from the original signal, a listener will not always perceive this error as inherently bad. Take for example a tube amplifier with "soft" clipping characteristics. By driving this sort of amp into clipping, the signal is compressed, but the added spectral components aren't nearly as annoying in comparison with those generated by "hard"-clipping amps like most solid-state designs. In fact, because the spectral envelope rapidly decays, comparatively, with increasing frequency, dynamics of transients appear perceptually more lively to many listeners, much more lifelike. Nothing wrong with that -- after all, we set up systems that we find pleasing musically. I suspect that the missing absolute dynamic range due to the effect of compression isn't noticeable to most listeners as the available audible dynamic range of their listening rooms isn't all that great either.
Virtually all home listening rooms cannot provide a dynamic range equal to that provided by a well-mastered CD, sad to say. Even taking into account the ability of the human ear to hear "into" the noise, in a fashion that decreases rapidly with increasing frequency, the dynamic range available to the listener will rarely exceed 90dB at 1kHz, and 70dB at 10kHz. Values of 80dB and 60dB respectively are far more common. Compression can often create a greater sense of perceived realism due to the raising of low-level components above the background noise floor of the listening environment, especially at low frequencies. Add to that the increased dynamics of transient attacks due to the handy nonlinear nature of tube amps and it's no wonder so many listeners to such systems routinely run their systems into compression, albeit unknowingly.
Once the listener learns to "hear" compression though, I would think the effect would then be so noticeable that one couldn't get around it any longer. Happened to me at least, many years ago. I've haven't been able to stand the sound of a compressing system ever since. Whether due to electrical signal compression or speaker compression, it makes no difference. Compression is compression.
Don't get me wrong, a nicely set up tubed system driven into slight compression has a lovely sound -- at times hypnotic. But after a while, it grates, at least to me. Your experience is of course mutually exclusive. I would never rank individual preference; that would be rude.
Nothing to gain
With all this in mind, one of the first things I set out to do when trying to put together a system that wouldn't compress was to consider the magnitude of the electrical signals in the chain. In so doing, I came to the inescapable conclusion that there is no need for any gain whatsoever in the preamp stage. This, of course, assumes that the sensitivity of the amplifier is such that a CD player, or any other line source, has an output sufficient to drive the amp to clipping. As an example, my Krell KSA-150 has an input sensitivity, unbalanced, of 1.8Vrms to achieve rated power. The onset of clipping occurs at 2.18Vrms, minimum. The output of a CD player, or DAC, is standardized at 2Vrms. Seems pretty clear to me that any gain would allow the amp to be driven to clipping without much effort. Doesn't sound like a good idea to me since the amp is solid state and there are no clipping indicators at all.
So the end result was that I abandoned active preamp stages altogether in favor of passive designs. But everyone says passives are well known for muting dynamics and reducing the perceived power in the bass. Why would the common wisdom arrive at this conclusion? My guess is that, in the case of the former, without compression, the typical listener is hearing the signal anew. Compression is likely so common that the lack of it is disconcerting. The problem of "missing" bass, I'm pretty sure, is related to the typically lower input impedance presented by most passive controllers to the source equipment interacting with the output coupling capacitor to raise the high-pass corner frequency of the unit's response. Of course, this would be most common with tubed output stages. But something just doesn't jibe here -- maybe it's just me.
Could it be that a combination of clipping-induced compression, the current fascination with "detail," and speakers that are often far from compression-free results in a system configuration that when presented with a non-clipped signal sounds simply dull and uninvolving? Just speculating here, but that's my guess. Of course, that and buck will get you a cup of coffee. It works for me though. What if a system were configured from the ground up to avoid clipping and compression? Would it then sound "live?" Depends on the room and recording, but I'd have to say yes. At least so far it seems that way. But I digress.
Initially, to get my system set up for passive use, the DAC had to be redesigned to take into account the minimum attenuation of the passive controller, currently an RLA unit that uses a variable-shunt/fixed-series type attenuator. Reworking the I-V stage to provide a full-scale output of about 2.5Vrms allowed the maximum input to the amp to reach 2.18Vrms, which is exactly what's necessary to avoid clipping. Now, the "volume" control can be turned to the stop without worry. The speakers are rated for 250W continuous music power, so they can't be overdriven either. This is basically a destruction-proof system. I like it this way.
One caveat though. The amp is designed so that it will act as a voltage-controlled voltage source (VCVS) into a load impedance as low as 1 ohm. This limitation is defined by the maximum output power of 150W into 8 ohms, which in turn defines the maximum allowable input voltage that could be applied through the passive controller and still maintain accurate output voltage into a speaker with a non-constant load impedance. In this specific case that would be 1.8Vrms. Given that the speaker presents a minimum impedance of roughly 2 ohms, a minimum attenuation of about 2dB would be advisable leaving about 1dB available for added "gain" when playing CDs, or any other high-level line source, where the full-scale range is not utilized. Using a potentiometer-type passive controller though will yield a full 3dB available excess "gain" owing to its minimum attenuation of 0dB. Obviously this isn't much of a defining characteristic when choosing between these two topologies.
With a claimed sensitivity of 95dB, the speakers can produce a decent acoustic output with little power yet are able to reproduce peaks without noticeable compression even when played at nearly deafening levels. Not state of the art, of course, but much better than most I've heard along the way. Interestingly, lowering the attenuation below 3dB is quite audible as a loss of upper-bass and lower-midrange definition, making the system sound overly present and bit bright if not oddly distorted. Naturally, this loss occurs pretty much exactly where the impedance of the speakers drops to a minimum.
Well, the numbers work out in theory, but that's never exactly all there is to it. Passive controllers have a deserved reputation for being finicky in their component matching. And justly so. Source output impedance, destination load impedance, attenuator input and output impedances, signal and ground switching, cable capacitance, etc. mandate careful attention in their use or you are just asking for trouble.
Figure 1a below depicts the simplest variable-shunt/fixed-series type passive controller setup; a variant of 1a but with an added discrete series resistor, 1b; a potentiometer-type passive controller, 1c; and a simplified active preamp, 1d. In Figure 2, these four configurations are represented by their equivalent circuits. All of these configurations are drawn as single-ended circuits without any reference to chassis connections or cable parasitics. Basically, these represent the differential circuit paths tangentially referred to in the previous article on cables.
Note that these are grossly simplified schematics. The resistances depicted exhibit in reality nonlinearities as do the capacitances from whence distortion may arise. The source and active preamp output resistances would be more accurately described as being in series with an inductance, which in turn would be paralleled with a resistance if any feedback were used in the design of the output stages. The cables would, of course, exhibit inductance, both mutual and differential; however, this inductance is very small in magnitude and has virtually no effect on signal transfer, so it will be neglected in the following discussion.
The most oft-repeated myth is that a passive controller has to drive the entire signal path all the way to the amplifier, represented in these figures by the block labeled "Destination." It certainly sounds like a reasonable thing to say, doesn't it? Fortunately, it's just not true, at least not to the degree many would have you believe, except for the one particular topology depicted in Figures 1a and 2a.
The math that shows this proposition is a bit complex, but if you think of the source output voltage, represented as Vout, as slewing rapidly from zero with the rest of the circuit having no initial charge, then in the instant after the slewing begins, Ccable1 appears as a dead short and the instantaneous output current, Iout, is limited only by Rout. Note that at this instantaneous increment of time, Ccable2//Cin is also a dead short and until the voltage across Ccable1 starts to climb, no current will pass through Rseries, or R2, and into Ccable2//Cin. This particular example is, of course, not directly applicable to the circuit shown in Figure 1a.
If we can safely assume that Rout is always much smaller than Rseries, or R2, then even as the voltage across Ccable1 climbs, due to the current flowing into it, the current that is then forced to pass through either Rseries, or R2, and into Ccable2//Cin is much, much smaller than the initial instantaneous current passing through Rout. This holds true throughout the interval wherein the output voltage is slewing. Thus, in any of the setups depicted in Figure 1, with the exception of 1a, the transient load seen by the source is then essentially Ccable1. Rseries, or R2, acts to isolate the parallel combination of Ccable2 and Cin much like the active gain stage in the active preamp.
If, however, Rout is actually a more complex impedance due to the presence of feedback in the output stage and if this impedance is very low, the peak current this simplified explanation would have you imagine is much larger than it could be in reality as the output stage would likely begin to actively limit the current. Of course, the actual numbers are dependent on maximum output slew rate (which is a function of the amplifier bandwidth and peak signal voltage), actual maximum peak current output (before current limiting begins to cut in), and the actual load capacitance. Getting an active, feedback-controlled output stage with active current limiting to operate perfectly without compression or multiplicative-disturbance- (signal dependent gain changes caused by active current limiting) induced transient-response errors is neither a simple nor inexpensively realized design task. Its even more difficult than reading that last sentence. Trust me.
Anyway, any current limiting that acts progressively within the output stage, like in virtually all op-amps, will cause some compression, and it won't be very easy to measure either as it acts transiently within the loop of the amplifier and causes the forward gain, internal to the op-amp itself, to vary during the current-limiting interval. Ugly. Very ugly, actually, and responsible for more good old spectral contamination that's going to just get worse in every downstream nonlinear stage that the signal passes through.
This means that the source equipment has to have either a large enough build-out resistor in series with an open-loop output stage so as to limit peak current to a safe level or an active, feedback-controlled output stage without active current limiting, again with a discrete build-out resistor to protect the stage from damage. I took the latter path in the DAC-1 modifications with a BUF-04 and a 76.8-ohm build-out resistor. The BUF-04 has no output-stage current-limiting circuitry and will be destroyed if shorted to ground during operation. The build-out resistor provides device protection while minimizing bandwidth limitations as a function of capacitive loading, not to mention keeping the amplifier stable by isolating the capacitive load.
You may then conclude that the load impedance presented by a passive controller, either the type depicted in Figure 1b or 1c, to any given source is not substantially different from that observed when using an active preamp. Again, in terms of resistance, this may not be so, but it can be made to be, albeit with an increased passive-controller output impedance that will strongly limit the allowable capacitive loading of the passive. One tangential conclusion that can be reached at this point is that regardless of the type of preamp used in a given system, Ccable1 should be minimized, or in other words the interconnect between source and preamp should be minimized.
All of the foregoing discussion, however, does not apply directly to the topology depicted in Figure 1a. In this particular passive design, there is no discrete series resistor. The series element is just the output impedance of the source itself. In this particular case, the source must indeed drive the entire circuit all the way to the amplifier. The McCormack TLC is designed around this topology and may very well be the origin of the myth that all passive designs must drive the entire downstream circuit, but that's just conjecture on my part. Given an output impedance in the source equipment that is independent of signal and a downstream circuit that exhibits very low capacitance, this topology should work just as well as any of the others. It's just that it will be much more sensitive than the other topologies to parasitics in the load side circuit.
The previous discussion concerned itself with the load as seen by the source equipment but what about the output impedance of the "preamp" as seen by the destination equipment? Here we find the real problems with the use of passive controllers. Whereas the output impedance of an active preamp is independent of attenuation level, it is totally dependent upon it in passive controllers. In fact, when presented with long interconnects or ones with high capacitance, or an amplifier with high input capacitance, the output impedance of a passive controller can lead to problems with band limiting if the problem is severe enough.
For example, let's assume a passive of the type depicted in Figure 1c with a total, or end-to-end, Rpot value of 50k ohms. This value is chosen so that the input impedance of the passive is roughly equal to the 47k-ohm standard most active preamps present to the source. At half span (6dB attenuation), the output impedance of the passive is 12.5k ohms, which in concert with a 2m cable of high capacitance interconnect, say 200pF total, and an input capacitance at the amplifier of about 220pF, the corner frequency of the simple, first-order, low-pass filter is roughly 30kHz, which results in about a 1.6dB attenuation at 20kHz. A more reasonable potentiometer value would be 10k ohms and the resultant attenuation at 20kHz would then be about 0.1dB, which is essentially inaudible. But again, the source has to be able to drive the 10k-ohm load of the passive input to the full peak voltage of the source material without error, a decidedly more difficult task for some designs.
Above, it was mentioned that the output impedance of a passive controller is strongly dependent on attenuation setting. Interestingly, the relationship between these two parameters varies markedly between the two topologies depicted in Figure 1b (or 1a) and 1c. The variable-shunt/fixed-series type (Figures 1a and 1b) yields a much lower output impedance over a much wider range of attenuation, at the expense of much lower input impedance, than the potentiometer type (Figure 1c).
The output impedance of any variable-shunt/fixed-series topology is equal to the series element resistance times the attenuation. For an attenuation of 12dB then, the output impedance would be approximately 250 ohms given a series resistance of 1k ohms, as is found in the RLA design. The input impedance is roughly equal to the series element value up to -20dB or so, rising nonlinearly to two times the series element value by -6dB, and 11 times at about -1dB. More precisely, the input resistance is equal to the series element resistance divided by the quantity, one minus the attenuation, or Rin=Rseries/(1-A), such that A is between 0 and Amax. Amax is defined by the maximum value of Rshunt. Interestingly, Rout=A(1-A)Rin.
A potentiometer-type topology will have an input impedance equal to the pot's end-to-end value and will not vary over the attenuation range. However, its output impedance will rise faster with attenuation setting than will the output impedance of the previous example. At an attenuation setting of 12dB, a 10k-ohm potentiometer-type passive will have an output impedance of 1875 ohms, considerably higher than the previous example. More precisely, the output resistance is equal to the attenuation times the quantity, one minus the attenuation, times the end-to-end potentiometer resistance, or Rout=A(1-A)Rpot, such that A is between 0 and 1.
Obviously, the variable-shunt/fixed-series type can be configured by the designer's choice of element values to achieve a higher input impedance characteristic at the expense of the output impedance. The Audio Synthesis design uses this approach and achieves a nice balance of reasonably high input impedance and tolerably low output impedance, albeit with a greater dependence upon destination-side capacitive loading than the RLA for instance.
For the user, the choice between these two topologies will depend on the ability of the source(s) to drive a low load impedance as might be presented by a passive of the type shown in Figures 1a and 1b. Often, this can only be determined experimentally. But then some people like to play mix and match. It's part of the hobby, and you're welcome for a new reason to keep playing the game.
Anyway, to sum up the differences in output impedance characteristics between the topologies, the topology depicted in Figure 1a can never have an output impedance greater than Rout and is always equal to Rout//Rshunt. The topology depicted in Figure 1b, however, can never have an output impedance greater than Rout+Rseries and is always (Rout+Rseries)//Rshunt. The topology depicted in Figure 1c has an output impedance that peaks at roughly Rpot/4 when the midspan attenuation is reached and is below this value on either side of this attenuation level. And, of course, the output impedance of the active preamp depicted in Figure 1d is always Rout2.
There is another subtlety involved in the use of designs based on those shown in Figures 1a and 1b. That is the selection by the designer of the actual shunt-element type. A pot connected as a rheostat or a stepped attenuator are the two primary choices, but a photoconductive cell, such as a CdS cell coupled to either a filament lamp or an LED, is another somewhat more exotic design that is found in Melos products, I believe.
The problem with rheostat-connected pots is twofold. Firstly, the slider contact resistance generates noise, a tiny amount to be sure, but it is nonzero nonetheless. The variable-shunt/fixed-series topology attenuates this noise inversely with the signal attenuation, which is clear upon examination of the equivalent circuit using superposition in the network analysis. Hint: Place the noise source in series with the variable element. In comparison, the potentiometer topology sums the slider noise directly with the output signal independent of attenuation. Thus the variable-shunt/fixed-series topology has a very slight signal-to-noise ratio advantage over most of the attenuation range. Note that this noise is can also be vibration dependent although better pots obviously won't exhibit this problem to any measurable degree.
Secondly, the slider of any pot is essentially an array of point contacts at best, and as such, the sheet current in the resistive element must depart from ideal planar flow such that it may flow through the slider pickups. This leads to a very slight nonlinearity in the variable element's resistance and thus a bit of distortion. Interestingly, the resistance is also a function of signal amplitude, making it even more of a complex issue; in particular, the variable element's resistance will increase with increasing signal, the opposite of compression. This effect is also present in the potentiometer-type topology, but is far less pronounced as potentiometers are designed to operate into as close to an open circuit as possible. Thus the input impedance of the destination equipment should be as high as possible so as to avoid any hint of nonlinearity.
In both topologies, however, a stepped attenuator eliminates these issues entirely and in their place substitutes a loss of continuous adjustment. Unfortunately, the open mechanical nature of a stepped attenuator leads to a greater sensitivity to EMI pickup. Shielding such an animal is hardly a simple task, but there are off-the-shelf solutions available such as the TKD part and the precision audio attenuators from Shallco. The photoconductive cell approach yields continuous attenuation adjustment while simultaneously eliminating the problems associated with slider noise and contact distortion, albeit at the expense of a more complicated circuit and the need for electrical power to run the circuit. This ruins the whole elegance of a passive design as far as I'm concerned, but what the heck -- it's a neat idea. However, any optical noise in the light source, especially if it's signal dependent due to poor bypassing or whatever cause, will generate distortion and noise. You just can't seem to get away from error sources no matter how hard you try.
One last point of interest here. One of the more persistent myths in the audiophile community is that the variable-shunt/fixed-series topologies (Figure 1a and 1b) are somehow better than any other topology because the "signal" passes only through the fixed series element and not the variable shunt element, thus allowing the use of a cheap pot and a high-quality resistor. Supposedly this saves money while achieving perfection. Naturally it does no such thing. A cursory examination of Figures 2b and 2c will prove this beyond a shadow of a doubt. Thevenin tells us that, by following the flow of signal current, both resistors in a voltage divider contribute to the output impedance. This is rather unavoidable. But then there are those who think the "signal" wire should be silver and the "ground" wire should be copper since copper makes the best "ground." Obviously the concepts of "signal," current, and voltage are easily confused, but enough on that.
To summarize, we've seen that what works best for you in your system will depend on the source and destination equipment design. Ideally, a source with a low output impedance and no current limiting, other than that provided by a discrete build-out resistor, will work well with any type of passive controller. Other types of source equipment may not work particularly well if they require a high impedance load to attain their specified gain and bandwidth. In this instance, a high-impedance passive controller is called for. However, it will likely cause bandwidth problems of its own if the passive-to-amp interconnect is high capacitance and/or if the amplifier input capacitance is high. In any case though, a short interconnect between the source and the passive is always recommended. If the maximum output impedance of the passive is relatively low, however, a low capacitance passive-to-amplifier interconnect may not be necessary as it will likely have negligible impact on bandwidth in the system. The choice between a stepped attenuator and a continuous potentiometer-type attenuation element will likely be left up to the listener's preference as it would be difficult to ascertain the audibility of such differences directly from measurements in my opinion. Correlation and all that, dont ya know.
With all this in mind, I've decided to take the tack that a stepped attenuator configured as shown in Figure 1c should be optimal for use in my system. A good ladder attenuator is no more expensive than a very good pot, such as a Penny & Giles RF15, and careful design of the chassis and wiring should minimize the EMI issues, what little there are likely to be. I don't think I'll miss the continuous attenuation characteristics of a pot given 2dB attenuation steps, but I think it'll be fun to find out. At least all the problems associated with a pot will be eliminated including the well known channel mistracking at low levels which is quite important to many listeners.
In process, then, is a replacement design for my RLA passive that will use an Arn Roatcap-designed 10k stereo stepped ladder attenuator. A particularly odd custom-made ELMA Type-04 rotary switch that maintains continuous connection of the input return paths with that of the output during selection switch rotation yet maintains isolation between the unselected input returns when the selector switch position is static will work better, in my estimation, than that of a bussed series of toggle switches as is done in the RLA design. This method will allow the actual, physically realized circuit to be exactly as shown in Figure 1c, only doubled for two channels of control as a pure dual-mono design.
The bussed toggle-switch selection method, while quite clever in how it addresses record output and monitor input connections, presents a significant problem in that it absolutely mandates bussing all the signal input returns together with the output. Otherwise there will common-mode transients during selection switching and the creation of a floating return path at the output when all inputs are deselected for muting. Can you say CLICK and bye-bye tweeters? No thanks. Even if this isn't a real problem in a given system, bussing all the returns together is a bad thing, as was discovered in the previous article discussing the complex common-mode nature of unbalanced interconnects. This was shown to be a common-mode noise problem in terms of induced current forced to flow between interconnected chassis, which exhibit differing chassis potentials, and because I have no need for a record loop, abandoning this design method is then of little consequence. The only potential drawback of the new design is that the toggle-switch method minimizes loop area at the connection points while the method required to terminate input cables on a rotary switch does not. EMI and crosstalk will naturally increase but hopefully not to audible levels. Care must be taken to keep signals well separated, but this isn't terribly difficult to arrange.
Naturally, this is a full-mono design with no connections whatsoever to the chassis, which of course can be connected to earth via a separate wire. Simple, low-cost Cardas RCAs will be used to provide a contiguous return path connection at the RCA shell and low-cost Canare GS6 miniature audio coax will be used for wiring. I'll report on the audible characteristics of this design versus those of the RLA design, if there are any of course, in a future column. Overall cost should be within the realm of most hobbyists and a complete schematic and parts list will be provided in the follow-up article.
As an aside, dealing with a passive at the system level should be considered before laying out hard cash. Ideally, all the sources' outputs should be earth referenced through connection to their respective chassis, either directly or through a parallel RC network, and the amplifier inputs should be floating but with an earthed chassis for safety. Conversely, if all the sources' outputs are floating and the amplifier input is ground referenced, the situation is identical. If neither of these are true in a given system, then an unbalanced passive controller will likely cause more problems than it solves. Serious problems will arise when there are ground referenced sources, and/or floating sources, and a ground referenced amp input. Hum city. I suppose it's also possible to have all the sources' outputs and the amp inputs floating while providing a single-point connection to earth at the passive controller, but I've never seen such a setup.
Without a system-level approach to the problem, it is often best instead to make use of an active preamp that has true differential inputs, even for unbalanced signals. This way, all the ground-loop problems are eliminated and higher-frequency common-mode noise is suppressed if the product is designed correctly. The tradeoff, of course, is the addition of noise and distortion in the signal path. While many would debate the audibility of such, I'd just as soon avoid the problem altogether and save some money at the same time. Cheap is good, especially when it gets you closer to the music. Who could ask for more than that?
Copyright © 1998 SoundStage!
All Rights Reserved