The TS350R1 builds on my experience from the prototype and is a practical & working design that fits in a standard extruded aluminum case. When powered from a +13-16V DC source it will deliver a low distortion 60Hz sine wave output of 350 VA continuous at 110 VAC in ambient temperatures up to about +40C. In boost mode it will supply up to 400 VA (input voltage & thermally limited). Likewise it handles inrush currents & transients and will tolerate a dead short on the output.
It will work with input voltages down to about 11V DC and automatically derates the output. It includes a front-panel LED meter to indicate the relative load and fault codes. Protections include a fused input, under/over-volt, over-current / overload, and overheat. The AC output is fully isolated. It is cooled by internal, temperature controlled fans.
Anderson PowerPole jacks are used to connect input power. Like some of my other projects, this one includes a micro-USB service port.
Specifications (as pictured)
Input voltage: 11-16 volts DC
Operating temperature: -20C - +40C
Maximum continuous input current: 45 amps
Idle current: 420mA
Output voltage: 100-120 volts AC RMS (110V AC RMS nominal)
This project is done. I'll continue to post logs discussing design details, challenges, problems, etc. I'll also add instructions detailing the inductor and transformer builds. and other important assembly information. Schematics & PCBs, design workbooks, and firmware are posted in the files section.
If you do build it please read thru all the instructions first. I recommend that you build from the input section forward: power -> start -> controller -> inverter -> transformer & supplies -> sine.
CAUTION: high voltages, currents, and temperatures exist in this design.
There is a problem with the layout of the current sense circuit that distorts the ‘B’ side signal. Here is a trace:
Channel 1 is the sense signal at the gate driver card pin connecting to the main board. Channel 3 is the ‘A’ side drive signal, so when 3 is low the sense signal is coming from the ‘B’ side of the bridge. Amazingly, it doesn’t appear to effect the safe operation of the unit. Below is a trace of over-current limiting prompted by a surge event on the sine output:
Channel 1 is the gate drive for the ‘A’ side and channel 4 is ‘B’. Channel 2 is the current sense signal from the gate card pin. Note the delay and subsequent higher DC for the ‘B’ side; but still short enough to protect the transistors.
Instruction ten describes a fix to improve the B signal and it’s highly recommended that you do this. Below is the output from the sense pin with the modification.
So what happened? Poor layout is to blame: the circulating currents in the bridge traces interfere with the signal. Using discrete wires avoids this.
A proper fix requires placing the amplifier closer to the sense resistors and at the same time keeping the distance from the amplifier output to the controller short as well. As explained in the log on the general design problem that is essentially a redesign.
The TS350R1 employs a discrete, high-side current sense circuit in the inverter primary that is the basis for all over-current protection & monitoring in the unit. Reference the main & inverter gate drive schematics for details.
The current signal is derived from a pair of 3mΩ (R302, R303) sense resistors that create a 1.5mV/Amp voltage drop across them. A resistor solution was chosen due to its compact size and low cost. They are placed in a high-side configuration due to the high currents and to provide better protection for the bridge. Low-side circuits can’t detect shorts occurring upstream and can contribute to ground bounce related problems for the gate drivers.
The challenge with high-side sensing is that it isn’t ground referenced so the conditioning circuit must operate from an floating supply or be capable of handling the common mode voltage. Since isolation isn’t practical the TI OPA197 (U2) is used. It allows for rail-rail inputs and supports slew rates up to 20V/uS which is satisfactory for per-cycle current limiting. It is configured as a differential amplifier. R4 & R5 in conjunction with D2 & D3 provide input protection from voltage spikes. R4 & R6 set the gain at 10 and Q1 level shifts the output (removes the CM voltage) for direct consumption by controller U101.
The controller processes the current sense signal to provide per-cycle [SOA] protection for the bridge MOSFETs and to perform overall power management thru two distinct hardware pathways using on-chip peripherals.
MOSFET SOA Protection Per-cycle protection uses DAC1 to provide a reference voltage to comparator CM2. The reference voltage is set to a value high enough (140A) to support the significant currents required of equipment startup surges but still within the MOSFET’s Safe Operating Area. CM2 compares the sense value to this reference and produces a H->L transition that is gated thru logic cell CLC3 to a Falling Event input of the COG (Complementary Output Generator). A transition immediately terminates the gate drive for that side of the bridge for that cycle.
The role of CLC3, along with PWM11, is to provide a blanking function at the beginning of each half cycle. Blanking prevents a premature termination of the gate drive and is set to a few hundred nano-seconds (defined by the DC of PWM11). This is necessary because of the inherent noise in the current sense signal. Noise is a problem for all sense circuits based on shunt resistors but it is more significant in high current, high frequency applications due to parasitic inductance and poor signal to noise ratios.
Current Averaging The current sense signal is also connected to the non-inverting input of opamp OPA2 which is configured as a voltage follower. This is necessary because of the relatively high impedance of the current sense circuit’s output and, most importantly, for OPA2’s tristate feature. Tristate, along with C1 (C102 on the main schematic) form a sample & hold circuit that averages the current sense signal from both sides of the bridge and allows the controller to derive RMS current. CLC2 gates OPA2’s output to ensure that it is only passing a signal when the bridge is active.
The voltage read from C1 represents an averaged value of the sense signal but it reads high. This is because the sense circuit delivers an exaggerated slope with a large spike (parasitics) at the end of each cycle which inflates the measured S&H voltage. Since the controller has limited processing bandwidth a linear coefficient is applied to reduce the read current to obtain a value that is +/- 1.5A of the actual RMS value.
The controller multiplies this RMS current value with the inverter’s input voltage to estimate the input power and uses this to perform load management. The power value drives the front panel LED meter and overload shutdown.
At the core of all power supplies is a transformer. It is central to the design and dictates the form and function of everything around it. A home brew design imposes additional requirements. Without specialized tooling and winding materials (e.g. Litz or foil wire, etc.) the design must also be constructible. Given this, my criteria were that the transformer must be:
From earlier work I knew that a center tap secondary is not practical due to the voltage gradient and turn-turn capacitance.
In a center tap secondary, the voltage measured across the legs is twice the output voltage. So a 200V secondary actually has 400V across the entire winding. In high frequency transformers this, combined with the turn-turn capacitance becomes a serious problem that results in significant ringing. Here’s an example:
Note the cursor value and the scale for channel 3 - 100V! Professionals get around this with lower capacitance magnet wire and special winding techniques that can only be done by machine (e.g. Bank).
Which is a shame since a center tap requires only two rectifiers instead of four. That may not seem significant but consider the consequences: space, dissipation, materials. In this design if a CT were viable I could have ditched the heatsink and likely used two SMD diodes saving space & halving the dissipation.
Decision 1: single winding secondary.
A center tap primary is possible and that would allow the use of a half-bridge - again saving significant space and halving dissipation. But there are drawbacks:
Window utilization is poor - each half winding is only doing work half time.
A larger core size would likely be required given the size of the winding - 2 turns of 20 AWG 10 strands becomes 4 turns.
Terminating the center tap would be problematic.
Greater chance of staircase saturation: any imbalance in each half winding (due to construction) creates a flux imbalance which over successive cycles walks the core into saturation.
Decision 2: single winding primary.
An inverter acting as an AC power source must be capable of handling large currents and supporting surge power well beyond their continuous rating. To do this the transformer must have a high power density and this requires operating it across the full BH loop in quadrants 1 & 3.
This is related to the single vs. center-tap primary discussion but emphasizes the power handling requirement of the transformer. To operate the transformer in this manner requires either a half or full bridge; but the prior discussion already discounted the half bridge due to manufacture concerns.
Decision 3: full bridge.
Most switching power supplies regulate thru some form of modulation - either the switching frequency, duty cycle, or both. Without a center tap secondary the conventional approach to regulation via the use of an output inductor to prevent the transformer core from resetting during off time isn’t practical. A current fed primary topology is a possibility but adds inductors & switches, introduces more losses, complicates the design, requires more space.
Decision 4: no regulation.
These decisions support the design criteria: a design that is manufacturable, simple, reliable, and (theoretically) efficient. The transformer can be operated at a high, constant, duty cycle, utilizing the full BH loop and achieving good power density. The complexity of a control loop is eliminated. But this shifts some burden to the rest of the design, both in front of and behind the transformer:
A full bridge will be used, doubling the space requirements and losses.
A full bridge rectifier is required, also doubling the space requirements and losses.
The secondaries will track the input voltage according to their turns ratio and both limit the...
I did a poor job on the systems engineering for this project. I spent a lot of time on specific problems like the layout of the inverter bridge, rectifier switching losses, heat sinks, firmware, etc. but I was too myopic. I didn’t bring it all together early enough in the design and this resulted in two major deficiencies.
Thermal management. I spent much time modeling heatsinks’ thermal performance, tab insulator materials, mounting height considerations to minimize parasitics and - key - the coefficients of each transistor on the dual sinks. But I stopped there: I compartmentalized the thermal models and did not establish the coefficients of adjacent areas or consider the entire system given hot air would be circulating in an enclosure.
I should have mocked-up the full layout with all key heat sources to establish coefficients and determine the size fans it would need. The as-built dissipates ~ 68 watts of heat at full load. Had I better established this early on I would have changed fan positions and sizing or maybe the entire design. I was fortunate that I had a convenient spot, and the power budget on the auxiliary supply, to add a third fan.
An early design goal was, aside from board size, for the unit to be enclosure agnostic. That’s how the pull/push arrangement for the fans on the board came to be. I realize now that in a design dealing with this much dissipation that the enclosure and the gear going in it are inextricably linked. An ah-ha related to this is that a fan must be positioned at the enclosure to ambient interface to ensure air is forced in or out of the unit. Otherwise internal fans will only be moving hot air around.
I also might have thought twice about trying to get this much power out of it. A two layer, two ounce PCB is woefully inadequate to handle this much current. Traces turn into copper pours, grabbing more space and reducing layout efficiency. Sometimes both layers must be used wasting even more. In spite of a very generous layout for the inverter primary it is dissipating almost five watts. That might not sound like much against the total losses but it is concentrated around other high dissipators and contributes significant heat stress.
AC Voltage Regulation. The sine controller has eight DDS sine patterns stored in flash and selects one based on the transformer’s auxiliary supply voltage (a proxy for the HV supply). It does a good job of keeping the no / light load voltage between 115 - 120V AC RMS. Under heavy loads the output tends to droop towards 100V. Most electronic devices won’t care but small motorized appliances (like a Dremel or mixer) will.
A few things contribute to this. Transformer regulation is poor; but this can be compensated for by adding a few turns to the secondary. Second, the DDS pattern imposes a voltage margin on the HV supply (you can see this in the Excel model). Essentially, the output voltage from the sine pattern on any given cycle is:
Vhv * Tdc = Vout
Where Vhv=High voltage supply Tdc = Duty Cycle, percent Vout = output voltage
The sine bridge on times are highest at the sine peak with a low input voltage. The DDS patterns are designed to ensure that there is a discernible off / on time for the same side high / low pair respectively to avoid high switching losses and maintain bootstrap capacitor charge. It also reduces the chance of violating transistor SOA and a resulting failure should a short or transient occur at just the right time.
Twelve volts is a ridiculous standard. Most of industry, sans automotive, abandoned (or never adopted) it more than a half century ago. Automotive just didn’t have an incentive and so the consumer world and all the stuff in its orbit are stuck with it. Hybrid & EVs are changing this but not soon enough.
Conceptually an inverter is simple: boost the voltage and modulate it in some way to obtain the desired AC output. There are several ways to do this depending on the application but all have in common the necessity of high currents if the input voltage is to be low.
A first order analysis is instructive. The inverter is to deliver 350 watts at 120 V AC RMS, 60Hz. Assume that only constant impedance, non-reactive (power factor of 1), loads will be used. The RMS current is:
The TS350r1 is a two-stage inverter. The first stage boosts the DC input voltage to greater than 200V DC depending on the input. The second stage uses Direct Digital Synthesis (DDS) to generate a 60Hz sine wave between 100 and 120V AC RMS. The AC output is fully isolated from the inverter and input.
The unit is connected to a power source via Anderson PowerPole 45A PCB mounted connectors. The (-) connects directly to inverter ground. The (+) is connected to two parallel wired 30A MINI fuses (F201, F202) providing 60A protection for the unit. The output of this parallel pair is connected to the input of a Solid State Relay (SSR) and a low power SPST on/off switch (S201). The output tap of the switch is connected to a 1.6A fuse (F203) powering the start and inverter auxiliary power supply circuit.
Start / Auxiliary Supply
The start circuit provides over-volt protection, inrush limiting, and a stable supply voltage for the inverter logic & gate drive circuitry. Two layers of over-volt protection are implemented. The first is auto-resetting. Q205 is a pass transistor that must be on for the auxiliary supply (and hence the unit) to function. In normal operation it is biased on via Q206 & R205. D203, R204, and Q203 form a voltage sensitive switch that, depending on tolerances, turns on (saturates) around 16.2V. When this occurs the gate of Q206 is pulled low and it turns off. In turn, R206 removes the negative bias from the gate of pass transistor Q205 and the auxiliary supply is turned off.
The second layer is a crowbar circuit formed by D202 and F203. D202's value is set so that it enters full conduction by ~ 30V. The excessive current flow opens F203 and unit repair is limited to fuse (and possibly D202's) replacement.
Inrush limiting functions thru the action of D204, R210 and the SSR. During a normal power-on sequence the SSR remains off and energy reaches the inverter's significant bulk capacitance via D204 and R210. A short in the inverter may open fuse F203; but if not R210 is a fusible resistor that is non-flammable. During normal operation the SSR is turned on by the controller prior to starting the inverter. D204 is necessary to prevent the auxiliary supply receiving power via the SSR once it is on and the power is switched off.
U201 is the auxiliary supply and consists of a Commercial Off The Shelf (COTS) DC-DC isolated SMPS capable of delivering a regulated 12V 500mA output. This powers the gate drives, fans, and 5V logic LDO, U202.
Solid State Relay
A discrete, single pole, SSR is implemented for inrush management and to avoid the need for an expensive, bulky, switch capable of handling the full input current. Q202 & Q204 implement the relay for the positive supply. Gate bias is provided by a simple isolated flyback circuit. It delivers a floating ~ 12V DC drive to the gates using a fixed duty cycle PWM provided by the controller. Q201 switches the primary and C201 provides bulk capacitance. Output recification & filtering is provided by D201 & C202.
The inverter operates at 50kHz as a DC transformer using a fixed on time (95% duty cycle) and a full bridge drive circuit for the transformer. Monolithic half-bridge gate drivers utilizing bootstrap circuits for the high side floating supply are used. High-side current sensing is provided by R302 and R303. The differential voltage across them is amplified by U2 and level-shifted by Q1 for processing by the controller. The controller can terminate the gate drive per-cycle and also derives RMS current & power.
The transformer consists of a single, 2-turn primary, 32-turn High Voltage (HV) secondary, and 3-turn auxiliary secondary. To reduce Common Mode (CM) noise two faraday screens between the primary & HV secondary are included: one...
This instruction is for building the transformer. It describes materials and fabrication guidance but assumes the reader has experience assembling magnetics. Additional information can be found in the magnetics zip file and the 'inductors' tab of the Excel design file.
The following materials are required:
TDK RM14 Power bobbin, mfg #B65888C1512T001, Digikey 495-5341-ND
High voltage secondary, N=32 consisting of two layers of 16 turns each, single strand 20 AWG
Primary, N=2, single layer, 10 strands 20 AWG
Auxiliary secondary, N=3, single layer, 1 strand 24 AWG
1) Wind high voltage secondary first. Two layers of 16 turns each. Terminate to pins 7 & 8. Apply one layer of tape over first layer. Begin second layer using foldback technique.
Above: 1st layer complete with tape.
Foldback taped in position and ready to wind second layer.
Second layer complete.
2) Apply one layer of tape to completed secondary. Leave unterminated (stuff into bobbin center) until complete with other windings.
3) Install Faraday screen (terminates to secondary (-)). The copper tape should be trimmed to a width of 17mm. Measure a length that allows for a few mm of overlap. The ground lead will connect to pin 12 on the bobbin so the turn should start on the opposite side of the bobbin.
4) Before completing the turn apply tape over the start of the turn to avoid a short. Then complete the turn.
5) Opposite the termination (other side of bobbin) solder a ground lead (24 AWG magnet wire) to the center of the screen and terminate to pin 12. Then apply a layer of tape: be sure to overlap the edges of the bobbin so the next screen doesn't short to it.
6) Install the second screen. This one will terminate to the primary (-) so the turn should begin on the secondary side. There is no pin available to connect the ground lead (it terminates to PCB when installed) so protect the lead by stuffing in bobbin when done. Otherwise the build is same as first one.
7) Winding the primary is difficult because it consists of ten individual strands of 20 AWG terminated to sets of pins. Pins 1-3 are for one end of the winding and 4-6 the other. Start by measuring ten strands to allow for two turns (with extra). Terminate 3 leads respectively to pins 1-3. One pin will need to accommodate four.
8) Wind two turns. Keep the strands in each turn as parallel as possible and cover as much of the bobbin as possible (important to minimize leakage inductance).
9) Terminate the strands to pins 4-6.
10) Then wind the three turn, 24 AWG auxiliary secondary on top of the primary. No tape is necessary and placement of turns is not critical. Terminate to pins 10 & 11.
Then apply a few layers of tape to secure & protect the windings.
11) Install the core and protect the primary faraday screen lead until ready for installation.
12) Use a DVM to check for shorts. No winding should be connected to another or the faraday screens.
13) Functional check. It's a good idea to make sure there are no shorted turns, particularly with the screens. This also provides a static validation of proper layering and taping.
For this test you will need a 2-channel oscilloscope, function generator, and a 1/4W resistor between 3-5KΩ. Connect the primary according to this diagram (image taken from McLyman, C. (2011). Transformer and Inductor Design Handbook (4th ed) New York CRC Press, pp 17-9.):
Leave the secondaries and screens open. Set the function generator to 50kHz generating a 4V P-P sine wave.
Sweep thru 400kHz. As 400kHz is approached the amplitude will increase and the phase will align with channel 1. Amplitude should peak within 10kHz of 400kHz. This is the transformer's resonant frequency and is defined by winding inductance and parasitic (lumped) capacitances.
IF you get no signal on channel 2 it means there's a shorted turn. The transformer will not work at all since the flux is shorted as well (it also means you'd have a dead short across the bridge).
If the resonant frequency is substantially different (esp. lower) it indicates a potential problem with the build.
Output Inductor Build
This instruction is for building the output inductor (two required - L401, L402). It describes materials and fabrication guidance but assumes the reader has experience assembling magnetics. Additional information can be found in the magnetics zip file and the 'inductors' tab of the Excel design file.
The following materials are required:
Magnetics High Flux toroid core, #C058547A2
Magnetics toroid mount, #TVH38134A
3M 1mil Polyimide tape 1205 (various widths) or equivalent
22 AWG MW35C HY magnet wire
You will likely need to acquire the core & mount from a Magnetics Inc. distributor such as Dexter or Elna.
1) The 'inductors' tab of the Excel design workbook provides the detailed winding data. The inductor consists of 100 turns of 22 AWG magnet wire (two strands). The winding will require about 2-1/2 layers and tape can be used on each layer to secure.
2) The precision of each layer and amount of tape use isn't critical.
3) The strands of each end should be terminated to the respective posts at each end of the mount.
4) Secure the wound inductor to the mount using a dab of high temp silicon RTV or similar.
5) Functional check. Like the transformer, it's a good idea to check the resonant frequency.
Same setup: 2-channel oscilloscope, function generator, and a 1/4W resistor between 3-5KΩ. Connect according to this diagram (image taken from McLyman, C. (2011). Transformer and Inductor Design Handbook (4th ed) New York CRC Press, pp 17-9.):
Resonant frequency should be around 300kHz. If the frequency is substantially lower then there's excessive capacitance and that could cause problems. In general, higher frequencies are okay.
No channel 2 signal indicates a short.
SSR Flyback Inductor
This instruction is for building the SSR flyback inductor - L201. It describes materials and fabrication guidance but assumes the reader has experience assembling magnetics. Additional information can be found in the 'ssr-flyback' tab of the Excel design file.
The primary consists of 6 turns and the secondary 17. 26 AWG wire is recommended for both for durability but the secondary can use 28 or even 30 AWG.
Polarity is critical: for the flyback to function the primary & secondary polarities must be opposite.
1) Wind the primary first. N=6. Space the turns evenly to cover the entire core.
2) Wind secondary, N=17. Start the winding between where the two primary leads exit the core. Wind a half-secondary to the left & right of it (one side will have 8 turns, the other 9) so that the winding is complete with the secondary leads exiting opposite the primary.
3) The inductor is wound correctly if it appears like the picture below. Note that the lead in the lower right exits the core's bottom and that in the upper right from the top. Those on the left half do the same but in reverse.
4) The function check for this one is to install it. Be sure to install with the pri & sec on the correct sides.
When the controller has the SSR on you should measure 12-15V DC across the G-S of either SSR MOSFET. Make sure you're measuring across and not ground referenced.