Jupiter 6 Waveshaper

Original Roland Chip

Introduction The Roland Jupiter 6 (and MKS-80) use a custom hybrid chip (EHM-S226W83S ) for transforming the CEM3340 oscillator waveforms to a consistent level of 5v centered around 0V (-2,5V to +2.5V). The CEM3340 creates waveforms that are at different levels (Pulse 0 to +12V, Triangle 0 to +5V, Sawtooth 0 to +10V) which need level shifting for mixing into the VCF. This is implemented by a mixture of SMD resistors and transistors.

The chip also contains a transistor based sync circuit, as Roland chose not to use the inbuilt CEM3340 sync (weak and strong), possibly because they wanted to control the sync on/off with a +5V programmable signal from the micro-controller and didn’t want use an analog switch.

Roland took this approach to reduce the footprint of the waveform translation to a minimum using a 12-pin SIP package. This enabled the Jupiter 6 PCB’s to be more compact than the Jupiter 8. An alternative approach is to use resistors and Op Amps but this is 5x the PCB size in the days before SMD components could be used.

Roland EHM-S226W83S

Pin Out The EHM-S226W83S pin out can be easily determined from the schematic in the Jupiter 6 Service Manual:

1.  +15V
2. -15V
3. Pulse Output
4. Pulse Input
5. Sync Input from second VCO sawtooth
6. Sync ON/OFF from microcontroller
7. Sync Output to CEM3340
8. Triangle Input
9. Sawtooth Input
10. Triangle Output
11. Sawtooth Output
12. GND

Tightly Packed!

How it works There are four separate circuits; one for each waveform and one for the sync. The triangle wave generated by the CEM3340 is the 0V to +5V and is level shifted to -2.5V and +2.5V by a PNP transistor (Q1) and a resistor network (R2, R5, R6).

The 0 to +10V sawtooth wave is reduced and shifted to -2.5V to +2.5V using three resistors (R1, R3, R4).

The 0 to +12V pulse wave is reduced and shifted to -2.5V to +2.5V using a more complex circuit of six resistors, an external capacitor and a NPN transistor (R13 – R17 and Q4.). The SH-101 service manual uses a similar circuit without the final level shifting. So it is easy to determine what most of the values are without measurement.

The sync circuit is also more complex. The sawtooth wave from the source VCO is switched off by grounding the input to a NPN transistors (Q2), or going high to enable the Sync. The sawtooth is then pulse shaped by a NPN transistor, 2 diodes, a capacitor and 3 resistors, The narrow pulse resets the VCO integrator via its capacitor (Q2, Q3, C1, R7-R12), creating hard sync.

Tear Down For my Jupiter 6 VCO clone I wanted to replicate the EHM-S226W83S circuit, along with the RA3A resistor array connected to the CEM3340. The array resistor values are easy to determine from the schematics and values in the MKS-80 service manual. The Jupiter 6 makes use of all 6 resistors in the array, whilst the MKS-80 only uses five of them.

The EHM-S226W83S was harder to replicate, as although the schematics are in the manual, the values are not printed and are a mystery. I bought one in April 2018 and removed the plastic encapsulation with acetone (took 2 days of soaking), so that I could then measure the various printed on resistors using a multi meter, some components had to be desoldered to be measured.

Prototyping Once the values had been obtained the full VCO schematics were laid out using Eagle CAD. Roland has implemented the VCO with some changes to the original CEM3340 datasheet  which have also been captured.

A prototype PCB with THD components was ordered in April 2022. The AMSynths Jupiter 6 VCO will operate with 12V power rails, so the resistor values will need changing anyway, as the original uses 15V power rails. I plan to keep with THD components rather than building SMD replicas.

Moog 901A/B VCO – Part I

Moog 901A/B

Overview In December 2021 one of my customers asked me if I could build the 901A/B VCO’s as an AMSynths product. In the past I had rejected the idea as they are relatively unstable and I cant get anywhere close to a Behringer 921A/B price point. However I have a small stock of CA3019 diode array chips (and they are still available new), and there is space in my “Big Moog” to fit a set of 901A/B/B/B, so let’s have a go! I will also add the rare 901C module which creates a LFO from a 901B.

The early 901’s used diodes whilst the later versions used enclosed diode array chips. The stability issue is caused by the uncompensated exponential converter, which was never corrected (MOS-LAB have  produced 901 VCO’s with a modern expo). The limited pitch range is caused by using a unijunction transistor as the sawtooth core of the module, 3 octaves of reasonable scaling appears the limit. I will be measuring this once my design is completed.

Research and Design The 901A/B went through many schematic changes in its life to make it more reliable/stable, so I will base my replica on the later 1970 version, to hopefully capture the best ideas from Bob Moog. This later version is the one in Klaus Schulze’s 1969 IIIP “Big Moog”. The modules have been successfully cloned by various 5U/MU manufacturers, and of course Moog have re-issued them as exact replicas. I haven’t seen one squeezed into EuroRack but the clones do seem relatively easy to get working.

My design criteria is to follow the original Moog schematics and use the same components, except where I cant find them or I can chose a better part. I will improve the power supply design as that is clearly what Bob was trying to achieve but without access to modern components in 1965! I have squeezed the designs into 8HP EuroRack modules with 3 PCB layers. No SMD components will be used (except possibly in the power supply), all THD, so I can solder them up myself.

Moog 901A – 1969

Stable Power The power supplies of the original are -6V, -10V, +12V and +11.5V, with the 901A powering the tracking circuit of the 901B from its +12V supply. Precision power rails are very important to the front end of a VCO and I have used modern LDO regulators across the modules.

The -6V is generated by LM2991 LDO regulators from the EuroRack -12V supply and a LM2941 LDO is used to generate the +11.5V in the 901B. The +12V power supply is interconnected across the modules using the BUS connector – see discussion below. The -10V is delivered from a LM337 adjustable linear regulator.

Voltage Divider The 901A uses a resistor ladder to generate the initial frequency control voltages, with a set of 12 voltages from -5V to +6V. I have used a precision voltage source (REF02)  to provide +5.00V, which is then buffered and inverted by a LT1013 Op Amp. VCO transposition is therefore limited to 5 rather than 6 octaves, but I doubt I would ever transpose up 6 octaves!

Moog 901B – 1969

Semiconductors The transistors used in the later versions are 2N3392 and 2N4058, which I will use as they are easy to obtain. Switching to modern transistors with higher HFE needs changes in resistor values which I wish to avoid. The transistors need to be hand matched in some of the circuits, for example the exponential generator.

There are two selected transistors; the 2N2646 UJT at Q10 and the 2N4058 at Q7. These will be hand selected using the Moog calibration process of the correct voltage at Test Point A and the sine waveform symmetry. The 2N2646 is surprisingly easy to obtain new, 50 years after it was used in the Moog 901B’s!

The later 901A modules use two CA3019 diode array chips as temperature compensation (heated like the CA3046 in the Prodigy) and are an improvement over the early versions which had no temperature compensation at all. I have located a source of NOS CA3019 chips and ordered a batch. There are five 1N34A germanium diodes, which are fortunately still available. The unobtainable CL-1 regulator diode is replaced with a contemporary 1N458A, which is what Moog used in the Minimoog re-issue.

Resistors Whilst the original made use of carbon composite and carbon resistors, with precision resistors in the voltage ladder. I have used 1% metal resistors throughout, with 0.1% in the voltage ladder of the 901A and the tracking circuit in the 901B’s (to match the tracking across the three 901B’s).

Trimmers All trimmers have been upgraded from single turn in the Moog to multiturn Bourns. The later versions of the 901A have Scale, High and Low compensation trimmers, with the Mid compensation trimmer removed (as Moog did). I have added a Pulse Width trimmer to the 901A to get accurate pulse widths. The 901B has a Tracking trimmer, as well as Sawtooth offset, Triangle symmetry and Sine offset and symmetry.

Capacitors The timing capacitor values are critical to the octave matching of the 901B and have the following values;

• 10uF – low mode, not critical
• 200nF – use x2 100nF in parallel
• 100nF – use 100nF
• 50nF – use 2x 100nF in series
• 25nF – use 22nF and 3n3F in parallel
• 12nF – use 10nF and 2n2F in parallel

The first three octaves will be accurate and use hand matched Wima MKP2 polypropylene capacitors, although the upper two octaves will be out a bit, This can be corrected manually by the frequency pot and its unlikely I will be switching octaves live. These capacitors have 5mm lead spacing to get them to fit on my 901A panel PCB.

There is a second set of capacitors on the  2P6W rotary switch that go into the Triangle output (to help symmetry?). These were independent of the timing capacitors in early 901A’s but later versions only added a capacitor at low and a high range points, and reused the timing capacitors for the first 4 octaves.

The rest of the capacitors are either ceramic (pF) or polyester (nF).

Mechanical Design The 901A and 901B are each built from three PCB’s, with the circuit spread across all available board areas.

• POT PCB holds the Alps rotary switch
• JACKS PCB holds the jacks sockets and pots.
• MAIN PCB holds the power supplies and 10-pin connector.

Traditional 1P12W and 2P6W rotary switches are used for the frequency ranges, they require more space between the PCB and panel than my usual designs. This created the 3 layer PCB approach which also provides more room for the circuits. The 901A and three 901B’s are inter-connected using a 6-pin IDC cable, this replicates the Moog design but with a more direct dedicated wiring system.

901C Output Stage This is a rare Moog module with only 15 made on perf board in 1967. It provides complementary outputs (one normal, one inverted), for each of the four waveforms and further 2 inversion modes for the pulse waveform. The circuit is transistor based and provides waveform amplification as well, with a dual gang gain potentiometer.

The circuit requires a 3P6W rotary switch which is expensive and simply too big to fit in an 8HP module. I have used CMOS 4051 switches to replicate the functionality, with a modern Alps rotary switch controlling the six signal paths on three layers (input, output+, output-). 2N3392 and 2N4058 transistors are used, along with metal film resistors. There is one ceramic capacitor and single turn trimmer to adjust the initial gain.

Behringer 1004 VCO

Behringer 1004

Introduction One of the projects I worked with Behringer on during 2019/20 was the range of ARP 2500 clone modules, and the toughest modules to get right were the 1004 VCO and 1050 Mix Sequencer. I bought a Behringer 1004 module in September 2021 to see how well the VCO had turned out and to consider using it in a 5U replica of the ARP 1045 Voltage Controlled Voice.

The Behringer 1004 The Behringer VCO is a close replica of the original ARP schematics and the early prototypes proved impressive and worked first time! SMD components are a big advantage when it comes to VCO accuracy and stability, the B2600 VCO’s also proved precise (possibly too precise). High quality SMD Tempco resistors are used in the VCO cores.

The Behringer Manchester team completed the production version and added LDO regulators to the VCO to provide well regulated +/-11.34V voltage rails, which helps VCO stability. This is the same approach I use at AMSynths to create precision voltage rails for VCO circuits. There are five 4580 dual Op Amps, which are low noise and use BJT’s not JFETS at the input. They are a good replacement for the original 1339 and 301’s.

Front of the PCB

Timing Capacitors The original ARP 1004 used a set of seven timing capacitors in the VCO cores which are way too big to fit in a Eurorack module, and required manual selection. Behringer have chosen 4x polyester capacitors for the larger values (yellow and grey boxes in the photo) and SMD ceramics for the lower values. The low range capacitors are larger than the original; 1,150nF, compared with 710nF.

What is surprising is that the 3.45:1 ratio of capacitor value between the triangle and sawtooth cores in the original has not been retained, it is more like 10:1. This probably explains the extra frequency scale trimmer for the Triangle core (VR5).

VR9 HI END ADJUST

Trimmers The B1004 has a total of six trimmers, five of which are blue multiturn versions and not accessible, and one single turn which is accessible by the customer from the rear of the PCB (VR9). The Quick Start manual provides no details about the trimmers, so here is what they do:

• VR10 OFFEST – adjust the overall frequency of the VCO
• VR11 SCALE – adjusts the overall volts per octave scale of the VCO
• VR12 SAW ADJUST – adjusts the shape of the sawtooth waveform
• VR5 TRI SCALE – adjust the volts per octave of the Triangle core
• VR6 TRI OFFSET – adjust the frequency of the Triangle core
• VR9 HI END ADJUST – adjust the scaling at higher frequencies (4-8KHz)

The additional triangle frequency trimmer is probably used to compensate for using standard capacitor values in the two cores and avoids capacitor selection.

Incorrect Sawtooth

Waveform Accuracy All the waveforms were accurate and calibrated correctly by the factory, except for the sawtooth. The SAW ADJUST trimmer adjusts the shape of this waveform, which is generated by the sawtooth VCO core. The triangle VCO core is separate and is used for generating sine and square/pulse waveforms. The sawtooth waveform that has not been calibrated will show a step in the waveform, but you are unlikely to hear the difference in the audio range. You will notice it when using the B1004 as a LFO.

My B1004 VCO showed exactly this issue and I removed the front panel and adjusted VR12 to create a linear sawtooth with a 10v p-t-p amplitude. I used an expensive oscilloscope to adjust the VCO but a free software plugin can be used, in fact I detected the issue using MScope running in Cubase. The pictures show the before and after, note also the corrected waveform amplitude.

Corrected Sawtooth

The pulse waveform has a nice wide range from 5% to 95% (in fact 1.2ms of 25ms at 40Hz, which is spot on), and is exactly the same as the original ARP specification. The waveforms are all 10V peak-to-peak, again matching the original specification. The Triangle and Sine are bi-polar +/-5V whilst the Sawtooth , Square and Pulse are unipolar 0 to +10V.

The 1004 uses 100R Tempco resistors which are at R1, R2 and R19. These are also used in the Behringer 1047 filter but in series to give 200R.

Frequency Accuracy Once warmed up for 20-30 minutes the B1004 proved to be calibrated accurately at the factory, with a good set of results from C1 to C4;

• C1 is -4 cents
• C2 is 0 cents
• C3 is +1 cent
• C4 is +3 cents
• C5 is +7 cents
• C6 is +11 cents

The VCO was factory tuned to C1, with the Coarse control at minimum and the Fine control at midway. The VCO goes sharp above 1kHz and the calibration notes of the original ARP 1004  suggest only 1 cent inaccuracy at 1, 2 and 4kHz – this is a pretty demanding specification. There is a High Frequency trimmer (HI END ADJUST) that may help achieve better scaling above 1 kHz, in my own tests this does not do a lot.

Conclusion The Behringer 1004 VCO has worked out very well, good waveform shape accuracy and pitch stability over 5 octaves. I am building one into a 5U 1045 replica, more details in a future post.