Mixer PSU - One
Reverse Engineering
〰️
Reverse Engineering 〰️
Top down image of the Numark M3 2-Channel Scratch Mixer “INPUT/OUTPUT PCB” for reverse engineering. The power supply section is on the left.
As of the last post, I have a broken mixer with identified culprits for repair. A pair of voltage regulators have failed to a short, killing the power supply. Admittedly there may be a downstream issue that has caused the magic smoke to be released, but a standard repair feels well within reach.
As the power supply design is fairly dated, it feels unaesthetic to repair identically. Presented with an opportunity to build something anew, I’ll take it. The previous design depended on a large line transformer and linear regulators - inefficient, hot, and while simple, somewhat crude. This existing sub-module then provides a reference design for a project. The sub-module has an input and output specification that can be (approximately) reverse engineered from inspecting the board and its components.
The I/O PCB underside. Power supply section is the thick traces on the left, while the right hand side is audio processing.
Inspection Method
To work out what is going on with this board, I take photos of the top and bottom sides. Entirely tracing out electrical nets is possible from just these images - it is a single-layer PCB (all electrical connections on the underside). There are no inner PCB layers to obfuscate how components are being linked together. Using an image editor I mirror the underside photo in the Y-axis, and add it as a semi-transparent layer to the top photo.
Superimposed images of both sides of the PCB, with manual tracing of nets and components.
Using new layers each time, I now trace over electrical nets in different colours. This is primarily to deduce the number of power rails and exactly how they are constructed. Components are identified and have their values taken, before being checked off with green crosses. From these traced nets I can now construct a schematic on pen and paper.
Schematic
Hand drawn schematic of the power supply circuit.
I am glad I traced out this circuit, rather than assuming its function. While it does have a fairly typical +/-12V & 5V rail setup, how it gets there is slightly odd to me. Firstly, the diode “bridge” as it seems from a first glance is actually not a bridge topology at all. Rather, two pairs of diodes team up with 1000uF capacitors to form a pair of opposite polarity half wave voltage doublers. Wikipedia provides a bit of insight into the circuit’s history and one can see we’re working with a design from 1913. Hats off for a real “if it ain’t broke” moment.
I had assumed the 10V label at the input refers to RMS, giving an amplitude of ~14V. Through two diode drops of 1V (standard 1N4004 rectifiers) and a half wave voltage doubler gives 26V at the 7812 input terminal. With a dropout voltage of 2V, the 14V of headroom seems slightly overkill for a 12V rail. This is mirrored in the -12V rail provided by 7912. Even the 5V rail is created from a simple half wave rectifier with output voltage ~13V into the 7805. Without adding component non-idealities it looks like the power supply will have an overall system efficiency around or below 30%.
Stepping the voltage all the way down to 10V only to have to double it again before regulation seems like too many unnecessary steps. Some limit (legal or otherwise) on the exposed AC voltage in this consumer electronics item is my best guess at a reason why. Otherwise one could start with 24V AC and do away with the doubling circuits entirely. Alternatively, the increased voltage dissipation required in the 5V regulator may have been too great. In turn, the comparatively lower current supply required of the higher voltage split rails could have led a designer to optimise for the lower voltage and higher current rail. Using circuits designed more recently than 1913 allows one to circumvent many of these considerations.
Output Specification
Maximum power dissipation figure from the Texas Instruments datasheet for this family of voltage regulators.
The output spec of each of the voltage rails will be limited by their regulators. Given this I can work out what I need in order to create a like-for-like (or better than) replacement for this module. The datasheet for these ICs provides a maximum power dissipation graph. The Numark implementation does not use heatsinks, and for the benefit of the doubt I’ll assume 25ºC ambient temperature (most likely higher inside the case, and design should be done for worst case). This leaves a power dissipation limit of approximately 3W per regulator.
From the previous section analysis, I estimate a voltage drop of 14V across the 7812 on the +12V rail. For power dissipation limit of 3W this leaves the circuit with just over 200mA of supply. I will generously round this up to 250mA, and assume the same for the -12V rail. The 7805 only(!) sees a voltage drop of approximately 8V, so with a similar calculation I’ll assume 500mA is sufficient on this rail.
But Wait!
All the input labels refer to the supply at the barrel jack as “10V AC 500mA”. This means the most power that could come into the mixer is 5W, and from the previous analysis the efficiency is at best 45%. Lets assume that 5W goes completely into each rail in turn, and never at the same time (a bonkers worst-case). In this case, providing 2.25W per rail is sufficient due to losses in the circuit being replaced. Again rounding currents up for safety margin gets:
+/-12V @ 188 -> 200mA each
+5V @ 450 -> 500mA
Super similar to the numbers concluded earlier, which adds confidence that this is a correct assessment of what is needed.
Other specifications of note such as ripple rejection and output noise voltage will be referenced later in the design. Don’t worry I have big (read; overkill but aesthetic) plans for these!
