Phono Preamplifier with Line Output Transformers

 

Having paid considerable amounts of money for a good turntable, an adequate  phono stage is essential to revive Maria Callas, Louis Armstrong, Ella Fitzgerald or other celebrated artists of days gone by. Many current preamplifiers are either lacking phono inputs or they are equipped with cheap phono stages, not meeting high-end requirements. However, separate audiophile phono stages are expensive and a DIY project may be an useful option.

click to enlarge

Fig.1: MK II phono stage. The switch at the side was added temporarily for listening tests concerning class A mode of the op amps (see text below).

Basic configuration

The phono stage is designed for MM and MC cartriges. JFETs, bipolars and op amps were chosen over tubes because transistors are suitable devices to handle very small signals in my humble opinion. Furthermore they allow a small enclosure that can be placed right next to the tonearm. Opamps improved dramatically during the last years and the old paradigm of a fundamental superiority of tubes or discrete solid state circuitry over opamps in phono stages has to questioned seriously (see related article by Junji Kimura).

A balanced design was chosen for several reasons:

  1. it provides common mode rejection and hence is much less sensitive for the picking up hum or RF,
  2. the headroom for input and output levels is doubled,
  3. DC drifts are cancelled out,
  4. even harmonics are cancelled out,
  5. the power supply load is kept constant relative to the music signal. The first stage's transistors as well as the op amps run at the same currents, but inverse in polarity.

On the other hand balanced designs have a signal / noise ratio 3db worse compared to similar single-ended design and the number of components is almost doubled. Furthermore balanced designs sometimes are considered being less precise than good single-ended designs. This is true if the parameters of the inverting an non-inverting stages are allowed to shift around are not forced to perfect symmetry. The circuitry of the phono stage ensures perfect symmetry by a really stable current source in the input stage, a floating RIAA network and precision op amps in the output stage.

Thinking about the basic circuit configuration I was in doubt whether to prefer active or passive RIAA equalization. I reviewed a number of schematics, questioned experts and finally decided in favor of a passive equalization (see comment on active versus passive RIAA). So the phono stage is simply built up by two linear, flat frequency response amplifying stages with the passive RIAA network between them (see phono stage schematic).

 

Input stage

The input stage is designed free from negative feedback (NFB). For real high resolution the input stage must be able to handle very small signals around 10-9 volts. With high amounts of NFB the input stage is confronted with the even much smaller voltage differences between non-inverting and inverting inputs. Therefore I am afraid to loose definition in a NFB loop within the input stage. On the other hand distortion is not a major concern having very small signals.

The input resistor depends on the manufacturers recommendation. For any MM cartrige the input resistor should be set to 47kOhm. For a Benz Micro MC Gold and a van den Hul Frog I set he input resistors to 510Ohm without capacitances.

The first stage was set up originally as an all-jFET differential amplifier with two cascoded pairs of low noise 2SK369 jFETs (Fig.2a). The cascode configuration provides a large bandwidth. It eliminates the so called Miller effect which is the capacitive feedback from drain to gate electrode of the jFET. This effect limits the upper frequency in a common-source circuit.

The 2SK369s were chosen because of their exceptional transconductance of 40mA/V that was not matched by any other low noise jFET I could find browsing though the data sheets (see 2SK369 datasheet). In general, the larger the transconductance figure for a device, the greater the gain it is capable of delivering, when all other factors are held constant.

The circuit provides enough slew rate and headroom to handle the boosted high frequency signals without any risk of overdrive.

However, in listening tests we found that bipolar transistors such as the BC550C (see datasheet)  as upper cascode transistors performed significantly better compared to jFETs (Fig.2b). In this circuit BJTs might be more ideal upper cascode devices due to their lower effective impedance. The upper device in a cascode should ideally hold the voltage across the device absolutely constant, regardless of current flow.

 

click for full schematic

Fig.2a:

All FET input stage

click for full schematic

Fig.2b:

FET / BJT input stage

Independent from jFETs or BJTs as upper transistors, the gain is 100 (40db) at the differential output (between a and b) and 50 from a or b to ground. This corresponds almost perfect to the theoretical gain according to the following equation.

          Gain = dID / dVGS x RD                                 (EQ 1) *

          dID = drain current swing

          dVGS = grain / gate voltage swing

          (dID / dVGS) = transconductance = 40mA/V for the 2SK369 (see datasheet)

          RD = drain resistor = 2.5kOhm

 * EQ 1 is a simplified approximation and is valid only if the dynamic resistance of the cascode configuration is much larger than RD

 

Bandwidth ranges from zero to 700kHz (-3db) (see gain vs. frequency plot). The balanced designs in figure 2a and b have the advantage of a doubled dynamic range compared to a single ended design and allows more than 100mVpp input voltage without before clipping. The schematic shows a non-balanced input variant of the circuitry. Real symmetry at the input can be achieved easily by slight modifications at the input.

Signal symmetry of the input stage is ensured by a cascoded current source (CS) with a low noise RF bipolar transistor BFR91 (see data sheet) and another 2SK369. The BFR91 can be considered a very high impedance source resistor of the jFET. The differential resistance of a cascoded CS is several GigaOhm and hence substantially higher than that of a normal jFET as a current source.

 The current source forces current swings of the two cascodes to perfect mirror images. Hence the differential input stage cancels out even harmonic distortions. Second harmonics are preferentially generated by FETs due to their square law responses. Another advantage of using a current source is the absolutely constant load of the power supply.

For some theoretical background see the article on cascodes on Nelson Pass’s homepage (www.passlabs.com) and the “special articles” on jFETs by Erno Borbely (www.borbelyaudio.com).

 

RIAA equalization

RIAA equalization is implemented by a floating low impedance network behind the differential input stage (Fig.3a). Virtually the frequency dependent components of the network are in parallel to the 2.5kOhm drain resistors and form frequency dependent loads at the drains (see schematic for illustration). Interestingly the network has not a single resistor in the direct signal path.

The equalization tailors the frequency response to fit that of the RIAA curve by varying the drain load and hence the amount of gain of the input stage according to the above equation EQ 1. Thus the RIAA network controls the first stage gain, but without any component in the primary signal path. Furthermore the input stage has less problems with slew rate limitations because the higher the frequency the lower the load and hence the amplitude (click here for illustration).

The passive equalization of the earlier MK I version of the phono stage (Fig.3b) looked quite similar but functioned entirely different:

The MK I version of the phono stage also had drain resistors of 2.5kOhm but a conventional voltage dividing network with relatively high impedance to avoid loading of the input stage (see complete MK I schematic).

After listening tests with the two different RIAA network designs we found the low impedance network performing significantly better compared to a high impedance network although the RIAA responses measured are almost identical (see RIAA plots).

click for full schematic

Fig.3a: RIAA network of new                  MK II phono stage

click for full schematic

Fig.3b: RIAA network of old              MK I phono stage

The RIAA reproduction standard specifies three equalization time constants, 3180µs, 318µs and 75µs. Theoretical RIAA response can be derived from the following equation:

Contrary to most commercial RIAA networks, a fourth time constant around 3.14µs was added to the network. The Neumann cutting amp manual states a roll off at about 50kHz to stop the treble boost and Neumann cutting lathes and amps are somehow Industry Standard. Therefore the commonly documented RIAA curve is incorrect, because that roll off is neglected (see comments on fourth time constant). In my RIAA I fitted an additional resistor in series with the small capacitor (time constant 75µs) to form a new time constant around 3.14µs to compensate for the roll off at 50 kHz (Fig.3a and b).

All RIAA capacitors are basically simple 10 % tolerance MKS types. I bought a bulk of thirty 100nF capacitors (which was around 10 Euro) and took the time to hand-match the capacitors between the two channels to an overall difference of less than 0.1 %. This is highly recommended, as stereo imaging relies on accurate channel balance. Please note, that cheap digital meters may be quite inaccurate with regard to absolute values, but are very useful to precisely match the relative values of capacitors.

The RIAA error remains within less than +/- 0.2db of the usual specification except around 20 kHz and above due to the added fourth time constant. All grey colored resistors in the schematic, including the drain resistors and the resistors at the non-invertig inputs of the op-amps are part of the RIAA network.

The 1µF coupling capacitors are Mundorf Mcap ZN caps. They also have to be matched carefully to achieve equal impedances and symmetrical roll off at the low frequency end.

 

Output stage

The second stage uses jFET-input precision op amps for further amplification of the signals tamed by the RIAA network. High input impedances are necessary to avoid divergences from the RIAA. Minimum offset is indispensable for (almost) DC-free primary windings of the line transformers. I chose OPA637 as a good compromise of low noise, high input impedance, high speed and minimum offset (see data sheet). The gain is set to 30 (30db) for MM cartriges. For MC cartriges I recommend a gain around 300 (50db).

For MM applications you can either use OPA637 or OPA627. The latter one is much easier to get over here in Germany. However, for MC applications you must use OPA637 because bandwidth of the OPA627 becomes insufficient at high gains (see Tab.1) due to its internal unity gain stability compensation. Swapping between OPA627 and OPA637 at gains around 50db is clearly audible which emphasizes Allen Wright’s (see www.vacuumstate.com) statement of a must of a minimum bandwidth at least one decade beyond the accepted audio range for sufficient phase accuracy.

 

Gain

OPA 627

OPA 637

30 db

400 kHz

> 1MHz

50 db

40 kHz

200 kHz

Tab.1: Theoretical bandwidth of OPA627 / 637 at different gain settings. 40kHz bandwidth is insufficient (also see Bode-Plot for theoretical gain vs. frequency).

 

The two jFETs at the outputs are constant current sources, set to a bias current of 2.5 mA. They force the (usually faster) npn output transistors of the op amps into class A operation which improved sound quality in blinded listening tests with independent persons. However, class A biasing is not recommended at the high gain setting (50db) of the OPA637, because the offset induced by the current sources may drive the output transformers into saturation.

In most cases the phono stage will be intended to drive unbalanced line stages or power amplifiers. Thus the inverted signals have to be recombined at the output, if you do not want to waste 50% if the signal energy. For several reasons a good audio transformer is the most elegant component to achieve this:

      • The transformer adds up the inverted signals and provides common mode rejection without adding any noise,
      • with the secondary coils paralleled the output impedances of the op amps are paralleled, which might be beneficial when driving longer cables.
      • Furthermore a transformer allows correct absolute phase of the phono signal, ground loops are broken up and high voltage shielding is provided to protect the expensive op amps.

The output impedance is 180 Ohm between 10Hz and 20kHz. Above 20kHz the impedance increases to 300Ohm at 100kHz (see plot of output impedances).

Generally it is unjustified to credit well manufactured audio transformers with all sorts of sonic ills. Studio designers treat transformers as basic components and prized recordings with great bass end and huge clarity could well have had three or four transformers in the signal path. The Lundahl LL1570XL’s were chosen after tests with different transformers (see www.lundahl.se). The LL1570XLs have frequency responses from 10Hz to 200kHz (+/- 0.5db) and self resonance points beyond 250kHz. The windings are electrostatically shielded (see data sheet) and achieve extremely low interwinding capacitances.

Please note that audio transformers with bifilar or quadfilar winding techniques have interwinding capacities of several nanofarads. These transformers are solely designed for 1:1 applications and are not suitable for step-up or step-down applications.

The RC combination at the secondary coils compensates for the transformer inductivity, the values were chosen for good square wave response. In need of a balanced output the secondary coils can be connected in series. The two 3.3M resistors shall prevent electrostatic charging between phono stage and line stage or power amplifier respectively.

 

Power supply

Textbooks or magazines tell that the power supply is responsible for 50 % of the sonic performance. Unfortunately common AC power supplies do not deliver pure 50Hz or 60Hz sine waves but carry lots of disturbances from computers, control impulses, switching power supplies, thyristor dimmers or motor controls and even communication. Expensive filters, varistors, chokes, oversized transformers or even isolating transformers are recommended as remedy. But even if the AC is filtered effectively, the problems may continue with voltage regulation or current peaks from the rectification.

A way not to solve but to eliminate the problems is the use of accumulators for noise free real DC. So power is generated by a stack of 9 volts NiMH accumulators to ensure 100 % real DC for handling the tiny signals from a MC cartrige.

Listening impressions

Comments and critics on audio circuits by the designer himself will always be subjective and heavily biased. However, from listening impressions of members of the Klangforum Dresden (see www.klangforum.de) with lots of high-end experience, the phonostage was considered being very detailed, with excellent sound-staging, good bass definition and especially a very natural reproduction of human voices. It seriously can compete with commercial high-end products.

Back to the top of the page

Home