Brian Lowe (Belleson) developed a unique custom designed transconductance amplifier (Bel Amp) with fully differential (balanced) low noise JFET inputs and current output for phono preamps. This article details the requirements and parameter values, while exploring design options. A second article explains the discrete op-amp design, complete with simulations and measurements.
The Starting Point
That first phono amplifier stage was a tube stage with cascode load and “button” battery negative grid bias (Figure 1). That became the input to my first RIAA phono preamp, which I used for many years. (Its ultimate demise was in a ditch on the side of a road in Texas, but that’s another story.) Over the next 40 plus years I designed and built a number of phono amps and preamps, some good, some not so good. During that time, in various jobs as an analog circuit designer, I learned a lot of related information. Recently, I was able to devote more time to another phonograph preamp design, one for commercial sale that would incorporate every possible circuit benefit I know [1].
The final solution uses a custom discrete op-amp that, as far as I know, is unique and will accommodate any moving coil or magnet cartridge load. My starting goals were straightforward, based on time-tested principles.
These goals are typical of quality op amps: low noise, low distortion, wide bandwidth, fast dynamic response, stable at unity gain, able to drive capacitive loads, a current limited output buffer, and made from easily available standard components. In development, an additional goal was added to make the high impedance VAS (voltage amplification stage) output available at a device pin. The reason will be revealed in part two of this article.
Noise
The key to a low-noise amplifier is the input amplification. Low resistor values are a given. More critically, noise at stage one gets amplified by every stage that follows, so lower input noise equals significantly lower output noise. Electrical noise comes in two types — current noise and voltage noise.
To minimize current noise, we use super low noise LSK389 junction field-effect transistors (JFETs) from Linear Systems. Low-frequency voltage noise (1/f and popcorn noise) in semiconductors originates from surface impurities, and JFETs are buried junction devices, thus they have inherently low 1/f noise. They are also voltage-controlled devices with picoamp input currents, keeping current noise very low at the amplifier inputs.
A typical op-amp front end uses a differential pair in a voltage amplification mode, as seen in Figure 2, which gives the benefit of cancellation of common mode noise of any sort, not just inherent thermal (Johnson) and flicker noise, but potentially outside noise influences, such as power supply fluctuations. For this application we chose the JFET differential pair for its combination of high input impedance and inherent low noise benefits.
The JFET “long-tailed pair,” also known as LTP, has lower gain than the equivalent bipolar circuit due to the lower transconductance of a JFET. Its transconductance can be increased by adding a PNP transistor [2] as seen in Figure 3. It is similar to the well-known Sziklai, or complementary feedback pair (CFP) typically configured as NPN with PNP, equivalent to a Darlington pair and used as a voltage follower. I suppose we could call this a CFP LTP.
This increases the transconductance (current out per voltage in) of the stage by a factor related to PNP hfe. However, this has a highly nonlinear transfer curve, and steady-state bias is dependent on nearly perfect device matching of hfe over its range of collector current, temperature, and process parameters. As a high gain input stage, it could possibly be used in an integrated circuit (with its inherent parameter matching) but not in a discrete amplifier build. One of the PNP transistors is almost certain to saturate and latch during start-up.
Input voltage changes modulate current through the input pair, resulting in output voltage changes across the load resistors. Base-to-emitter and gate-to-drain junction capacitance decreases proportionally to the square root of junction voltage. (Note that inherent base-to-emitter capacitance is higher than base-to-collector because doping levels are denser, making a narrower junction and thus thinner “dielectric.”) This capacitance modulation from the signal is, of course, frequency dependent, and for complex audio signals it creates distortion. The same effect occurs with active current source loads instead of resistor loads, except with higher load impedance the effect is more pronounced due to the resulting higher gain.
A more interesting approach is to add some resistors to make a complementary feedback amplifier (CFA) as in Figure 4, where the JFET drain resistor modulates the PNP base emitter voltage, which modulates emitter current from the PNP. Richard Marsh discussed a clever CFA in Linear Audio, Volume 3 [3] with voltage gain, not current, and was not differential.
Figure 4 works as follows:
- Assume I1=I2=I3/2 for DC bias (more discussion later)
- R1 and R2 provide VBE bias for PNP transistors.
- Re1 and Re2 are emitter resistors for PNP transistors and provide negative feedback.
- As input voltage to the JFET gates swings positive and negative with respect to each other, current flows left or right through RL, giving output voltage signal as amplified input.
- Stage gain (actually transconductance) without feedback from Re1 and Re2 is, to first order, the product of JFET gm, PNP gm and source current from I1.
- Input pair is fully differential.
Distortion
Distortion in amplifiers occurs due to a nonlinear transfer from input to output. It can happen anywhere in the signal chain. Elements with gain such as transistors are inherently nonlinear and are “linearized” using techniques such as feedback or special operating modes (e.g., the linear portion of a FET curve or valve/tube). Some circuit elements are often presumed and modeled as linear and yet are not. Resistors change value with temperature and voltage. Capacitors change with thermal, mechanical, and voltage variation. Inductors become nonlinear and/or saturate at high current and have (minor) resistance variation with temperature. A single op-amp RIAA playback circuit can induce amplitude distortion at treble frequencies when closed loop gain exceeds the op-amp’s gain within the audio frequency range. Distributing gain across multiple stages extends bandwidth per stage when using amplifiers with a fixed gain bandwidth product (the lower the gain, the higher the bandwidth).
Semiconductor junctions have capacitance that varies with applied voltage, and it cannot be eliminated [4, 5]. In fact, it is used intentionally in some circuits (e.g., radio tuners), to adjust oscillator frequency. In audio, the key phrase is “with applied voltage.” By keeping node voltage changes small, voltage induced distortion can be minimized.
One way to achieve this is using op-amps in inverting mode, keeping their input stage fixed at a “virtual ground.” Or as analog guru Jim Williams put it, “always invert, except when you can’t,” which is known as the Jim Williams rule [6]. The signal exists as a change in current, not voltage.
Another way to minimize distortion is with feedback. Higher forward path gain will, in a feedback circuit, null out distortion [7]. For a phono stage it’s not so simple because op-amps typically have decreasing open loop gain as frequency increases. Op-amp output stages are typically class AB push-pull. This architecture can be a source of cross-over distortion, which is independent of gain but can have frequency dependence.
Figure 5 shows a typical single stage RIAA circuit implementation. To prevent amplitude distortion, the open loop gain of the amplifier must exceed the required closed loop gain across the audio bandwidth. Using OPA1641, a low noise JFET input op-amp will not suffice for a moving coil amp that requires 60dB (1000x) or more of gain. Figure 6 shows open loop gain vs. frequency and, above 10kHz, gain is less than 60dB. Thus, in this circuit, with this op-amp, treble above 10kHz will not get amplified as required, causing distortion in, for example, vocal sibilance or cymbal crispness. Other problems with this design have also been described [8].
There are other op-amps with higher open loop gain (e.g., LME49710) that are better for single stage high gain requirements. Figure 7 illustrates open loop gain variations for different commercial op-amps from SPICE simulations. The models were obtained from manufacturer’s websites, so they are assumed to be accurate. The op-amp discussed in this article is shown as Top Amp (since then it was renamed as Bel Amp to avoid confusion with an existing professional product and prior trademark).
As mentioned, distortion can be ameliorated with negative feedback. However, using feedback around an op-amp means the negative input has an impedance that depends on the gain requirement, so it’s difficult to make a balanced input with input loading that’s independent of other circuit factors. Look again at Figure 5 and notice how only the “+” input can have its impedance set with no other connections. Phono cartridges, especially moving coil cartridges, work best with balanced adjustable impedance loading at the amplifier input. The only way to implement that is with two high impedance fully differential inputs.
Do the Math
Refer to Figure 4 for the following analysis of the CFA current stage. We want transconductance y, the change in output current versus input voltage:
Some Real Numbers
For DC steady state biasing, the LSK389A datasheet [9] shows a vgs vs. id transfer curve that’s most linear between -3 ≤ vgs ≤-1. This is the Id current range we want for least distortion. By choosing 4mA as I1 and I2, and 8mA for I3, we can split roughly 2mA between the two transistors in half of the CFA. Depending on values of Rd1 and Re1, JFET current ranges from Id=1mA to 2.8mA, and PNP from Ic=1.2mA to 3.1mA (see Table 1). Rd is selected as 649Ω to allow 2mA emitter current to give voltage across Re of 0.5V to 1V, depending on how the 4mA is split between the JFET and PNP.
Given these values, transconductance of the stage can be calculated from the LSK389A datasheet:
for half of the differential input stage. With Re=649Ω, negative feedback reduces the calculated stage transconductance gain. Using Re=0 allows calculation of maximum possible transconductance by reducing Equation 7 to:
Next Month
This completes Part 1 of the article. In Part 2, we will further develop the discrete op-amp by adding Folded Cascode and Transimpedance stages and an output buffer. Part 2 will conclude with simulations and measurements. aX
References
[1] Brilliance, Belleson, LLC, https://www.belleson.com/oc/products/phono-preamplifiers/brilliance-phono-preamplifier
[2] B. Cordell, “Linear Systems LSK489 Application Note,” Linear Systems, Inc., Figure 4 and Figure 8, www.linearsystems.com/applicationnotes/lsk489-app-note.
[3] R.N. Marsh, “A Headphone Buffer/Amplifier and Auto-EQ for Headphones,” Figure 2, p. 32, Linear Audio, Volume 3, April 2012.
[4] B. Cordell, Designing Audio Power Amplifiers, p. 368, Second Edition, Routledge, 2019.
[5] P. R. Gray, R. G. Meyer, Analysis and Design of Analog Integrated Circuits, p. 38, John Wiley & Sons, 1977.
[6] J. Williams, Analog Circuit Design, Art, Science and Personalities, p. 52, Butterworth-Heinemann, Stoneham MA, 1991.
[7] B. Putzeys, “The F-word or, why there is no such thing as too much feedback,” Linear Audio, Volume 1, www.linearaudio.net/index.php/f-word-or-why-there-no-such-thing-too-much-feedback-0
[8] B. Cordell, “VinylTrak – A full-featured MM/MC phono preamp”, Linear Audio, Volume 4, p. 134, September 2012.
[9] LSK389 A/B/C/D Ultra-Low Noise Monolithic Dual N-Channel JFET Amplifier, Linear Systems, www.linearsystems.com/_files/ugd/7e8069_c90834b817214602abcbf263293295f2.pdf
This article was originally published in audioXpress, February 2025
Source link