Fan

ZekEng Internship Project

Fan-controller reverse engineering, simulation, PCB redesign, humidity-sensor research, and embedded audio DSP development.

Българска версия / Original Bulgarian internship log

Project overview

The main goal of this internship was to analyze and redesign an existing fan-controller board that was prone to failure. The work covered reverse engineering, LTspice simulation, component selection, schematic capture, PCB layout, fabrication, assembly, and supporting research into humidity sensing.

While the redesigned boards were being manufactured, the project was expanded with an embedded-audio task based on the Electrosmith Daisy platform. This included real-time reverb, audio passthrough, SD-card access, WAV parsing, and initial convolution-reverb development.

The original reverse-engineering files are available in Files for Reverse Engineering.

Repository contents

• Fan/ — KiCad schematic, PCB layout, project backups, fabrication files, and schematic PDF
• LTSpice Files/ — LTspice tutorials, component tests, power-supply simulation, and fan-controller simulation
• Files for Reverse Engineering/ — source material for the original board
• HumSensorResearch/ — humidity-sensor material research and reference papers
• DaisyExamples/pod/ — Daisy Pod audio experiments and convolution project
• Images/ — project photos, plots, screenshots, and supporting media
• README_BG.md — the original Bulgarian day-by-day internship log

LTspice simulation and circuit analysis

The simulation work started with the LTspice workflow in LTSpice Files/TutorialRoman/TutorialRoman.asc, covering component placement, wiring, parameter assignment, directives, and signal analysis.

AC analysis

.ac dec 50 1k 10Meg

This sweeps the circuit from 1 kHz to 10 MHz, using 50 logarithmically spaced points per decade, to measure gain and phase response.

Transient analysis

.tran 0 300u 100u 0.01u

This evaluates voltages and currents over 300 µs, starts saving data after 100 µs, and limits the maximum timestep to 0.01 µs.

Fourier analysis

.four 500k 10V(out)

The transient output is decomposed into harmonics around a 500 kHz fundamental to assess signal quality.

Parameter sweep

.step lin param Rgain 100k 200k 20k

Rgain is swept from 100 kΩ to 200 kΩ in 20 kΩ increments, with the analyses repeated for every value.

Identified behavior

The transient simulation revealed a point at which the output departed from the expected sine wave. Cursor measurements showed that the operational amplifier required measurable time to overcome the combined VBE voltage drop of transistors Q1 and Q2. Of the six simulated cases, the third trace—R1 = 140 kΩ, with a 14 V amplitude—was the minimum acceptable result.

Imported models and component tests

The CD4000-series models and the MAC97A8 triac model were added to the LTspice project. Supporting tests include:

• NAND Tests.asc — CD4093B NAND-gate test with a 12 V-amplitude sine wave
• SW.asc — LTspice voltage-controlled switch experiments
• PMU.asc — power-supply simulation
• Ventilator.asc — complete fan-controller simulation

The simulated supply reaches approximately 11.3 V after 160 ms and then gradually rises to about 11.6 V within two seconds.

Fan-controller redesign

The board was recreated in KiCad and revised to improve reliability while retaining the original mechanical constraints of approximately 14.5 mm × 32.5 mm.

Schematic and component changes

• Replaced the original 220 nF capacitor with a suitably rated part that fits the board.
• Corrected an intermediate selection error where a 22 nF part had been chosen instead of 220 nF.
• Changed the resistor topology from parallel to series.
• Selected a new Zener diode because the original component could not be identified.
• Verified and corrected the MAC97A8 triac pinout.
• Reviewed pulse-current capability and physical dimensions when selecting replacement parts.

PCB layout and custom footprints

The PCB work included:

• A custom test point with a 1.8 mm pad diameter and 1.1 mm drill
• A revised humidity-sensor symbol and footprint
• Improved placement of high-voltage resistors R21–R24
• A custom 220 nF capacitor footprint supporting multiple package sizes and updated 3D models
• A custom footprint and 3D representation for two horizontally mounted 100 µF capacitors
• Component rearrangement so the fan connection pads are not hidden under the electrolytic capacitors

After review, Gerber fabrication files were exported to Fan/Gerbers.zip, and the prototype was sent to Hardtech for production. A bill of materials was also prepared, including cost estimates for prototype quantities of 25 boards and mass production at 1,000 boards.

Manufactured prototypes

The boards and components were received and three prototypes were assembled for evaluation.

Humidity-sensor research

The original resistive humidity sensor was coated with latex, which degraded when exposed to moisture. Several hydrophilic sensing materials were investigated:

• Polyimide, including the possibility of a flexible-PCB sensor
• Polyaniline (PANI), which was rejected because of its cost and excessively high resistance
• Polyvinyl alcohol (PVA)
• Carbon-based construction for improved durability

Measurements from the original sensor indicated a nominal resistance of approximately 1.8 MΩ, falling to around 1 MΩ at 100% relative humidity. A major obstacle was the lack of reliable resistance-versus-area data for candidate coatings at a defined thickness.

To avoid this uncertainty, a commercial HR202 humidity sensor was removed from another fan, evaluated, and selected for the redesign.

Reference material is stored in HumSensorResearch/, including Polyaniline_based_impedance_humidity_sen.pdf.

Daisy embedded-audio work

While waiting for PCB fabrication, an Electrosmith Daisy development board was configured using the official C++ development-environment guide.

The following Daisy Pod examples were developed:

• Reverb — generates a signal and applies reverb
• Passthrough — routes computer audio through the board to the headphone output without processing
• PassthroughReverb — applies reverb to the passthrough signal
• Convolution — WAV loading, parsing, and convolution-reverb development

Demonstrations:

• Blink
• Generated signal with reverb
• Audio passthrough
• Audio passthrough with reverb

Convolution and WAV processing

The convolution goal was to load an impulse-response WAV file from an SD card and reproduce the acoustic response on the incoming audio signal. The tunnel-entrance impulse response is stored under DaisyExamples/pod/Convolution/deprecated/mono files/.

The source file was prepared with SoX:

sox tunnel_entrance_f_1way_mono.wav -r 48000 -b 24 -c 1 tunnel_entrance_48k.wav gain -n -3 rate -v

This converts the file to the Daisy board's native 48 kHz sample rate, uses a mono 24-bit output, normalizes the peak to −3 dBFS, and enables SoX's high-quality resampling filter.

Serial logging fix

Back-to-back PrintLine calls caused serial output to stall. A short delay provided an initial workaround:

hw.DelayMs(1);

A proper Flush() method was then added to the libDaisy logger and the library was rebuilt:

template <LoggerDestination dest>
void Logger<dest>::Flush()
{
    if(tx_ptr_ > 0)
    {
        TransmitBuf();
    }
    tx_ptr_ = 0;
}

SD-card access and RIFF/WAV parsing

After resolving file-access issues, the firmware successfully opened the SD-card file and verified its bytes against a hex viewer.

A WAV file is a RIFF container, so the audio data chunk is not guaranteed to begin at byte 37. Tools may insert chunks such as LIST or bext before the audio payload. The parser was therefore changed to scan dynamically for both the fmt and data chunks instead of assuming fixed offsets.

A successfully parsed test file reported:

RIFF: RIFF
File size: 480072
WAVE: WAVE
fmt : fmt
Audio Format: 1
Channels: 1
Sample Rate: 48000
Byte Rate: 96000
Block Align: 2
Bits Per Sample: 16
data: data
Data Size: 480002
Data Offset: 78

The WAV handling code was separated from main, and FatFs FIL objects were moved to global storage after debugging initialization problems with local instances. A file-copy routine was implemented to read and rewrite both the WAV header and sample data, providing a way to validate the SD-card data before integrating it into the convolution pipeline.

Main outcomes

• Reverse-engineered and simulated the original fan-controller circuit
• Identified important transient behavior and power-supply characteristics
• Redesigned the schematic and compact PCB with corrected components and footprints
• Generated fabrication files, a BOM, and production cost estimates
• Assembled three physical prototypes
• Replaced an uncertain custom humidity-sensing coating with an HR202 sensor
• Built real-time Daisy audio examples and established SD-card WAV parsing for convolution work