An I/O controller for virtual pinball machines: accelerometer nudge sensing, analog plunger input, button input encoding, LedWiz compatible output controls, and more.

Dependencies:   mbed FastIO FastPWM USBDevice

Fork of Pinscape_Controller by Mike R

/media/uploads/mjr/pinscape_no_background_small_L7Miwr6.jpg

This is Version 2 of the Pinscape Controller, an I/O controller for virtual pinball machines. (You can find the old version 1 software here.) Pinscape is software for the KL25Z that turns the board into a full-featured I/O controller for virtual pinball, with support for accelerometer-based nudging, a real plunger, button inputs, and feedback device control.

In case you haven't heard of the concept before, a "virtual pinball machine" is basically a video pinball simulator that's built into a real pinball machine body. A TV monitor goes in place of the pinball playfield, and a second TV goes in the backbox to serve as the "backglass" display. A third smaller monitor can serve as the "DMD" (the Dot Matrix Display used for scoring on newer machines), or you can even install a real pinball plasma DMD. A computer is hidden inside the cabinet, running pinball emulation software that displays a life-sized playfield on the main TV. The cabinet has all of the usual buttons, too, so it not only looks like the real thing, but plays like it too. That's a picture of my own machine to the right. On the outside, it's built exactly like a real arcade pinball machine, with the same overall dimensions and all of the standard pinball cabinet hardware.

A few small companies build and sell complete, finished virtual pinball machines, but I think it's more fun as a DIY project. If you have some basic wood-working skills and know your way around PCs, you can build one from scratch. The computer part is just an ordinary Windows PC, and all of the pinball emulation can be built out of free, open-source software. In that spirit, the Pinscape Controller is an open-source software/hardware project that offers a no-compromises, all-in-one control center for all of the unique input/output needs of a virtual pinball cabinet. If you've been thinking about building one of these, but you're not sure how to connect a plunger, flipper buttons, lights, nudge sensor, and whatever else you can think of, this project might be just what you're looking for.

You can find much more information about DIY Pin Cab building in general in the Virtual Cabinet Forum on vpforums.org. Also visit my Pinscape Resources page for more about this project and other virtual pinball projects I'm working on.

Downloads

  • Pinscape Release Builds: This page has download links for all of the Pinscape software. To get started, install and run the Pinscape Config Tool on your Windows computer. It will lead you through the steps for installing the Pinscape firmware on the KL25Z.
  • Config Tool Source Code. The complete C# source code for the config tool. You don't need this to run the tool, but it's available if you want to customize anything or see how it works inside.

Documentation

The new Version 2 Build Guide is now complete! This new version aims to be a complete guide to building a virtual pinball machine, including not only the Pinscape elements but all of the basics, from sourcing parts to building all of the hardware.

You can also refer to the original Hardware Build Guide (PDF), but that's out of date now, since it refers to the old version 1 software, which was rather different (especially when it comes to configuration).

System Requirements

The new config tool requires a fairly up-to-date Microsoft .NET installation. If you use Windows Update to keep your system current, you should be fine. A modern version of Internet Explorer (IE) is required, even if you don't use it as your main browser, because the config tool uses some system components that Microsoft packages into the IE install set. I test with IE11, so that's known to work. IE8 doesn't work. IE9 and 10 are unknown at this point.

The Windows requirements are only for the config tool. The firmware doesn't care about anything on the Windows side, so if you can make do without the config tool, you can use almost any Windows setup.

Main Features

Plunger: The Pinscape Controller started out as a "mechanical plunger" controller: a device for attaching a real pinball plunger to the video game software so that you could launch the ball the natural way. This is still, of course, a central feature of the project. The software supports several types of sensors: a high-resolution optical sensor (which works by essentially taking pictures of the plunger as it moves); a slide potentionmeter (which determines the position via the changing electrical resistance in the pot); a quadrature sensor (which counts bars printed on a special guide rail that it moves along); and an IR distance sensor (which determines the position by sending pulses of light at the plunger and measuring the round-trip travel time). The Build Guide explains how to set up each type of sensor.

Nudging: The KL25Z (the little microcontroller that the software runs on) has a built-in accelerometer. The Pinscape software uses it to sense when you nudge the cabinet, and feeds the acceleration data to the pinball software on the PC. This turns physical nudges into virtual English on the ball. The accelerometer is quite sensitive and accurate, so we can measure the difference between little bumps and hard shoves, and everything in between. The result is natural and immersive.

Buttons: You can wire real pinball buttons to the KL25Z, and the software will translate the buttons into PC input. You have the option to map each button to a keyboard key or joystick button. You can wire up your flipper buttons, Magna Save buttons, Start button, coin slots, operator buttons, and whatever else you need.

Feedback devices: You can also attach "feedback devices" to the KL25Z. Feedback devices are things that create tactile, sound, and lighting effects in sync with the game action. The most popular PC pinball emulators know how to address a wide variety of these devices, and know how to match them to on-screen action in each virtual table. You just need an I/O controller that translates commands from the PC into electrical signals that turn the devices on and off. The Pinscape Controller can do that for you.

Expansion Boards

There are two main ways to run the Pinscape Controller: standalone, or using the "expansion boards".

In the basic standalone setup, you just need the KL25Z, plus whatever buttons, sensors, and feedback devices you want to attach to it. This mode lets you take advantage of everything the software can do, but for some features, you'll have to build some ad hoc external circuitry to interface external devices with the KL25Z. The Build Guide has detailed plans for exactly what you need to build.

The other option is the Pinscape Expansion Boards. The expansion boards are a companion project, which is also totally free and open-source, that provides Printed Circuit Board (PCB) layouts that are designed specifically to work with the Pinscape software. The PCB designs are in the widely used EAGLE format, which many PCB manufacturers can turn directly into physical boards for you. The expansion boards organize all of the external connections more neatly than on the standalone KL25Z, and they add all of the interface circuitry needed for all of the advanced software functions. The big thing they bring to the table is lots of high-power outputs. The boards provide a modular system that lets you add boards to add more outputs. If you opt for the basic core setup, you'll have enough outputs for all of the toys in a really well-equipped cabinet. If your ambitions go beyond merely well-equipped and run to the ridiculously extravagant, just add an extra board or two. The modular design also means that you can add to the system over time.

Expansion Board project page

Update notes

If you have a Pinscape V1 setup already installed, you should be able to switch to the new version pretty seamlessly. There are just a couple of things to be aware of.

First, the "configuration" procedure is completely different in the new version. Way better and way easier, but it's not what you're used to from V1. In V1, you had to edit the project source code and compile your own custom version of the program. No more! With V2, you simply install the standard, pre-compiled .bin file, and select options using the Pinscape Config Tool on Windows.

Second, if you're using the TSL1410R optical sensor for your plunger, there's a chance you'll need to boost your light source's brightness a little bit. The "shutter speed" is faster in this version, which means that it doesn't spend as much time collecting light per frame as before. The software actually does "auto exposure" adaptation on every frame, so the increased shutter speed really shouldn't bother it, but it does require a certain minimum level of contrast, which requires a certain minimal level of lighting. Check the plunger viewer in the setup tool if you have any problems; if the image looks totally dark, try increasing the light level to see if that helps.

New Features

V2 has numerous new features. Here are some of the highlights...

Dynamic configuration: as explained above, configuration is now handled through the Config Tool on Windows. It's no longer necessary to edit the source code or compile your own modified binary.

Improved plunger sensing: the software now reads the TSL1410R optical sensor about 15x faster than it did before. This allows reading the sensor at full resolution (400dpi), about 400 times per second. The faster frame rate makes a big difference in how accurately we can read the plunger position during the fast motion of a release, which allows for more precise position sensing and faster response. The differences aren't dramatic, since the sensing was already pretty good even with the slower V1 scan rate, but you might notice a little better precision in tricky skill shots.

Keyboard keys: button inputs can now be mapped to keyboard keys. The joystick button option is still available as well, of course. Keyboard keys have the advantage of being closer to universal for PC pinball software: some pinball software can be set up to take joystick input, but nearly all PC pinball emulators can take keyboard input, and nearly all of them use the same key mappings.

Local shift button: one physical button can be designed as the local shift button. This works like a Shift button on a keyboard, but with cabinet buttons. It allows each physical button on the cabinet to have two PC keys assigned, one normal and one shifted. Hold down the local shift button, then press another key, and the other key's shifted key mapping is sent to the PC. The shift button can have a regular key mapping of its own as well, so it can do double duty. The shift feature lets you access more functions without cluttering your cabinet with extra buttons. It's especially nice for less frequently used functions like adjusting the volume or activating night mode.

Night mode: the output controller has a new "night mode" option, which lets you turn off all of your noisy devices with a single button, switch, or PC command. You can designate individual ports as noisy or not. Night mode only disables the noisemakers, so you still get the benefit of your flashers, button lights, and other quiet devices. This lets you play late into the night without disturbing your housemates or neighbors.

Gamma correction: you can designate individual output ports for gamma correction. This adjusts the intensity level of an output to make it match the way the human eye perceives brightness, so that fades and color mixes look more natural in lighting devices. You can apply this to individual ports, so that it only affects ports that actually have lights of some kind attached.

IR Remote Control: the controller software can transmit and/or receive IR remote control commands if you attach appropriate parts (an IR LED to send, an IR sensor chip to receive). This can be used to turn on your TV(s) when the system powers on, if they don't turn on automatically, and for any other functions you can think of requiring IR send/receive capabilities. You can assign IR commands to cabinet buttons, so that pressing a button on your cabinet sends a remote control command from the attached IR LED, and you can have the controller generate virtual key presses on your PC in response to received IR commands. If you have the IR sensor attached, the system can use it to learn commands from your existing remotes.

Yet more USB fixes: I've been gradually finding and fixing USB bugs in the mbed library for months now. This version has all of the fixes of the last couple of releases, of course, plus some new ones. It also has a new "last resort" feature, since there always seems to be "just one more" USB bug. The last resort is that you can tell the device to automatically reboot itself if it loses the USB connection and can't restore it within a given time limit.

More Downloads

  • Custom VP builds: I created modified versions of Visual Pinball 9.9 and Physmod5 that you might want to use in combination with this controller. The modified versions have special handling for plunger calibration specific to the Pinscape Controller, as well as some enhancements to the nudge physics. If you're not using the plunger, you might still want it for the nudge improvements. The modified version also works with any other input controller, so you can get the enhanced nudging effects even if you're using a different plunger/nudge kit. The big change in the modified versions is a "filter" for accelerometer input that's designed to make the response to cabinet nudges more realistic. It also makes the response more subdued than in the standard VP, so it's not to everyone's taste. The downloads include both the updated executables and the source code changes, in case you want to merge the changes into your own custom version(s).

    Note! These features are now standard in the official VP releases, so you don't need my custom builds if you're using 9.9.1 or later and/or VP 10. I don't think there's any reason to use my versions instead of the latest official ones, and in fact I'd encourage you to use the official releases since they're more up to date, but I'm leaving my builds available just in case. In the official versions, look for the checkbox "Enable Nudge Filter" in the Keys preferences dialog. My custom versions don't include that checkbox; they just enable the filter unconditionally.
  • Output circuit shopping list: This is a saved shopping cart at mouser.com with the parts needed to build one copy of the high-power output circuit for the LedWiz emulator feature, for use with the standalone KL25Z (that is, without the expansion boards). The quantities in the cart are for one output channel, so if you want N outputs, simply multiply the quantities by the N, with one exception: you only need one ULN2803 transistor array chip for each eight output circuits. If you're using the expansion boards, you won't need any of this, since the boards provide their own high-power outputs.
  • Cary Owens' optical sensor housing: A 3D-printable design for a housing/mounting bracket for the optical plunger sensor, designed by Cary Owens. This makes it easy to mount the sensor.
  • Lemming77's potentiometer mounting bracket and shooter rod connecter: Sketchup designs for 3D-printable parts for mounting a slide potentiometer as the plunger sensor. These were designed for a particular slide potentiometer that used to be available from an Aliexpress.com seller but is no longer listed. You can probably use this design as a starting point for other similar devices; just check the dimensions before committing the design to plastic.

Copyright and License

The Pinscape firmware is copyright 2014, 2021 by Michael J Roberts. It's released under an MIT open-source license. See License.

Warning to VirtuaPin Kit Owners

This software isn't designed as a replacement for the VirtuaPin plunger kit's firmware. If you bought the VirtuaPin kit, I recommend that you don't install this software. The VirtuaPin kit uses the same KL25Z microcontroller that Pinscape uses, but the rest of its hardware is different and incompatible. In particular, the Pinscape firmware doesn't include support for the IR proximity sensor used in the VirtuaPin plunger kit, so you won't be able to use your plunger device with the Pinscape firmware. In addition, the VirtuaPin setup uses a different set of GPIO pins for the button inputs from the Pinscape defaults, so if you do install the Pinscape firmware, you'll have to go into the Config Tool and reassign all of the buttons to match the VirtuaPin wiring.

TCD1103/TCD1103.h

Committer:
mjr
Date:
12 months ago
Revision:
109:310ac82cbbee
Parent:
104:6e06e0f4b476

File content as of revision 109:310ac82cbbee:

// Toshiba TCD1103 linear CCD image sensor, 1x1500 pixels.
//
// This sensor is conceptually similar to the TAOS TSL1410R (the original 
// Pinscape sensor!).  Like the TSL1410R, it has a linear array of optical
// sensor pixels that convert incident photons into electrical charge, an
// internal shift register connected to the pixel file that acts as an 
// electronic shutter, and a serial interface that clocks the pixels out
// to the host in analog voltage level format.
//
// The big physical difference between this sensor and the old TAOS sensors
// is the size.  The TAOS sensors were (by some miracle) approximately the
// same size as the plunger travel range, so we were able to take "contact"
// images without any optics, by placing the plunger close to the sensor,
// back-lighting it, and essentially taking a picture of its shadow.  The
// Toshiba sensor, in contrast, has a pixel window that's only 8mm long, so
// the contact image approach won't work.  Instead, we have to use a lens
// to focus a reduced image (about 1:10 scale) on the sensor.  That makes
// the physical setup more complex, but it has the great advantage that we
// get a focused image.  The shadow was always fuzzy in  the old contact 
// image approach, which reduced the effective resolution when determining 
// the plunger position.  With a focused image, we can get single-pixel 
// resolution.  With this Toshiba sensor's 1500 pixels, that's about 500 
// dpi, which beats every other sensor we've come up with.
//
// The electronic interface to this sensor is similar to the TAOS, but it
// has enough differences that we can't share the same code base.
//
// As with the 1410R, we have to use DMA for the ADC transfers in order
// to keep up with the high data rate without overloading the KL25Z CPU.
// With the 1410R, we're able to use the ADC itself as the clock source,
// by running the ADC in continous mode and using its "sample ready" signal
// to trigger the DMA transfer.  We used this to generate the external clock
// signal for the sensor by "linking" the ADC's DMA channel to another pair
// of DMA channels that generated the clock up/down signal each time an ADC
// sample completed.  This strategy won't work with the Toshiba sensor,
// though, because the Toshiba sensor's timing sequence requires *two* clock
// pulses per pixel.  I can't come up with a way to accomplish that with the
// linked-DMA approach.  Instead, we'll have to generate a true clock signal
// for the sensor, and drive the DMA conversions off of that clock.
//
// The obvious (and, as far as I can tell, only) way to generate the clock
// signal with the KL25Z at the high frequency required is to use a TPM -
// the KL25Z module that drives PWM outputs.  TPM channels are designed
// precisely for this kind of work, so this is the right approach in terms of
// suitability, but it has the downside that TPM units are an extremely scarce
// resource on the KL25Z.  We only have three of them to work with.  Luckily,
// the rest of the Pinscape software only requires two of them: one for the
// IR transmitter (which uses a TPM channel to generate the 41-48 kHz carrier
// wave used by nearly all consumer IR remotes), and one for the TLC5940
// driver (which uses it to generate the grayscale clock signal).  Note that
// we also use PWM channels for feedback device output ports, but those don't
// have any dependency on the TPM period - they'll work with whatever period
// the underlying TPM is set to use.  So the feedback output ports can all
// happily use free channels on TPM units claimed by any of the dedicated
// users (IR, TLC5940, and us).
//
// But what do we do about the 2:1 ratio between master clock pulses and ADC
// samples?  The "right" way would be to allocate a second TPM unit to
// generate a second clock signal at half the frequency of the master clock, 
// and use that as the ADC trigger.  But as we just said, we only have three 
// TPM units in the whole system, and two of them are already claimed for 
// other uses, so we only have one unit available for our use here.
//
// Fortunately, we can make do with one TPM unit, by taking advantage of a 
// feature/quirk of the KL25Z ADC.  The quirk lets us take ADC samples at
// exactly half of the master clock rate, in perfect sync.  The trick is to
// pick a combination of master clock rate and ADC sample mode such that the
// ADC conversion time is *almost but not quite* twice as long as the master
// clock rate.  With that combination of timings, we can trigger the ADC
// from the TPM, and we'll get an ADC sample on exactly every other tick of
// the master clock.  The reason this works is that the KL25Z ADC ignores
// hardware triggers (the TPM trigger is a hardware trigger) that occur when
// a conversion is already in progress.  So if the ADC sampling time is more
// than one master clock period, the ADC will always be busy one clock tick
// after a sample starts, so it'll ignore that first clock tick.  But as 
// long as the sampling time is less than *two* master clock periods, the
// ADC will always be ready again on the second tick.  So we'll get one ADC
// sample for every two master clock ticks, exactly as we need.
//
// This is all possible because the ADC timing is deterministic, and runs on
// the same clock as the TPM.  The KL25Z Subfamily Reference Manual explains
// how to calculate the ADC conversion time for a given combination of mode
// bits.  So we just have to pick an ADC mode, calculate its conversion time,
// and then select a TPM period that's slightly more than 1/2 of the ADC
// conversion time.
//
//
// Pixel output signal
//
// The pixel output signal from this sensor is an analog voltage level.  It's
// inverted from the brightness: higher brightness is represented by lower
// voltage.  The dynamic range is only about 1V - dark pixels read at about 
// 2V, and saturated pixels read at about 1V.  
//
//
// Inverted logic signals
//
// The Toshiba data sheet recommends buffering the logic signal inputs from 
// an MCU through a 74HC04 inverter, because the sensor's logic gates have
// relatively high input capacitance that an MCU might not be able to drive 
// fast enough directly to keep up with the sensor's timing requirements.  
// SH in particular might be a problem because of its 150pF capacitance,
// which implies about a 2us rise/fall time if driven directly by KL25Z
// GPIOs, which is too slow.
//
// The software willo work with or without the logic inversion, in case anyone
// wants to try implementing it with direct GPIO drive (not recommended) or 
// with a non-inverting buffer in place of the 74HC04.  Simply instantiate the
// class with the 'invertedLogicGates' template parameter set to false to use 
// non-inverted logic.
//
//
// How to connect to the KL25Z
//
// Follow the "typical drive circuit" presented in the Toshiba data sheet.
// They leave some of the parts unspecified, so here are the specific values
// we used for our reference implementation:
//
//   - 3.3V power supply
//   - 74HC04N hex inverter for the logic gate inputs (fM, SH, ICG)
//   - 0.1uF ceramic + 10uF electrolytic decoupling capacitors (GND to Vcc))
//   - BC212A PNP transistor for the output drive (OS), with:
//     - 150 ohm resistor on the base
//     - 150 ohm resistor between collector and GND
//     - 2.2K ohm resistor between emitter and Vcc
//

#include "config.h"
#include "NewPwm.h"
#include "AltAnalogIn.h"
#include "SimpleDMA.h"
#include "DMAChannels.h"


template<bool invertedLogicGates> class TCD1103
{
public:
    TCD1103(PinName fmPin, PinName osPin, PinName icgPin, PinName shPin) :
       fm(fmPin, invertedLogicGates),
       os(osPin, false, 6, 1),    // single sample, 6-cycle long sampling mode, no averaging
       icg(icgPin), 
       sh(shPin),
       os_dma(DMAch_TDC_ADC)
    {
        // Idle conditions: SH low, ICG high.
        sh = logicLow;
        icg = logicHigh;

        // Set a zero minimum integration time by default.  Note that tIntMin
        // has no effect when it's less than the absolute minimum, which is
        // the pixel transfer time for one frame (around 3ms).  tIntMin only
        // kicks in when it goes above that absolute minimum, at which point
        // we'll wait for any additional time needed to reach tIntMin before
        // starting the next integration cycle.
        tIntMin = 0;

        // Calibrate the ADC for best accuracy
        os.calibrate();
        
        // ADC sample conversion time.  This must be calculated based on the
        // combination of parameters selected for the os() initializer above.
        // See the KL25 Sub-Family Reference Manual, section 28.4.4.5, for the
        // formula.  We operate in single-sample mode, so when you read the
        // Reference Manual tables, the sample time value to use is the
        // "First or Single" value.
        const float ADC_TIME = 2.1041667e-6f; // 6-cycle long sampling, no averaging

        // Set the TPM cycle time to satisfy our timing constraints:
        // 
        //   Tm + epsilon1 < A < 2*Tm - epsilon2
        //
        // where A is the ADC conversion time and Tm is the master clock
        // period, and the epsilons provide a margin of safety for any 
        // non-deterministic component to the timing of A and Tm.  The
        // epsilons could be zero if the timing of the ADC is perfectly
        // deterministic; this must be determined empirically.
        //
        // The most conservative solution would be to make epsilon as large
        // as possible, which means bisecting the time window by making
        // A = 1.5*T, or, equivalently, T = A/1.5 (the latter form being more 
        // useful because T is the free variable here, as we can only control
        // A to the extent that we can choose the ADC parameters).
        //
        // But we'd also like to make T as short as possible while maintaining
        // reliable operation.  Shorter T yields a higher frame rate, and we
        // want the frame rate to be as high as possible so that we can track
        // fast plunger motion accurately.  Empirically, we can get reliable
        // results by using half of the ADC time plus a small buffer time.
        //
        fm.getUnit()->period(masterClockPeriod = ADC_TIME/2 + 0.25e-6f);
        
        // Start the master clock running with a 50% duty cycle
        fm.write(0.5f);

        // Allocate our double pixel buffers.  
        pix1 = new uint8_t[nPixAlo * 2];
        pix2 = pix1 + nPixAlo;
        
        // put the first DMA transfer into the first buffer (pix1)
        pixDMA = 0;
        clientOwnsStablePix = false;

        // start the sample timer with an arbitrary epoch of "now"
        t.start();

        // Set up the ADC transfer DMA channel.  This channel transfers
        // the current analog sampling result from the ADC output register
        // to our pixel array.
        os.initDMA(&os_dma);

        // Register an interrupt callback so that we're notified when
        // the last ADC transfer completes.
        os_dma.attach(this, &TCD1103::transferDone);
        
        // Set up the ADC to trigger on the master clock's TPM channel
        os.setTriggerTPM(fm.getUnitNum());
        
        // clear the timing statistics        
        totalXferTime = 0.0; 
        maxXferTime = 0;
        minXferTime = 0xffffffff;
        nRuns = 0;

        // start the first transfer
        startTransfer();
    }
        
    // logic gate levels, based on whether or not the logic gate connections
    // in the hardware are buffered through inverters
    static const int logicLow = invertedLogicGates ? 1 : 0;
    static const bool logicHigh = invertedLogicGates ? 0 : 1;
    
    // ready to read
    bool ready() { return clientOwnsStablePix; }
        
    // Get the stable pixel array.  This is the image array from the
    // previous capture.  It remains valid until the next startCapture()
    // call, at which point this buffer will be reused for the new capture.
    void getPix(uint8_t * &pix, uint32_t &t)
    {
        // Return the pixel array that ISN'T assigned to the DMA.
        if (pixDMA)
        {
            // DMA owns pix2, so the stable array is pix1
            pix = pix1;
            t = t1;
        }
        else
        {
            // DMA owns pix1, so the stable array is pix2
            pix = pix2;
            t = t2;
        }
    }
    
    // release the client's pixel buffer
    void releasePix() { clientOwnsStablePix = false; }
    
    // figure the average scan time from the running totals
    uint32_t getAvgScanTime() { return static_cast<uint32_t>(totalXferTime / nRuns);}

    // Set the requested minimum integration time.  If this is less than the
    // sensor's physical minimum time, the physical minimum applies.
    virtual void setMinIntTime(uint32_t us)
    {
        tIntMin = us;
    }
    
protected:
    // Start an image capture from the sensor.  Waits the previous
    // capture to finish if it's still running, then starts a new one
    // and returns immediately.  The new capture proceeds asynchronously 
    // via DMA hardware transfer, so the client can continue with other 
    // processing during the capture.
    void startTransfer()
    {
        // if we own the stable buffer, swap buffers
        if (!clientOwnsStablePix)
        {
            // swap buffers
            pixDMA ^= 1;
            
            // release the prior DMA buffer to the client
            clientOwnsStablePix = true;
        }
        
        // figure our destination buffer
        uint8_t *dst = pixDMA ? pix2 : pix1;
        
        // Set up the active pixel array as the destination buffer for 
        // the ADC DMA channel. 
        os_dma.destination(dst, true);
        
        // Start the read cycle by sending the ICG/SH pulse sequence
        uint32_t tNewInt = gen_SH_ICG_pulse(true);

        // Set the timestamp for the current active buffer.  The ICG/SH
        // gymnastics we just did transferred the CCD pixels into the sensor's
        // internal shift register and reset the pixels, starting a new
        // integration cycle.  So the pixels we just shifted started
        // integrating the *last* time we did that, which we recorded as
        // tInt at the time.  The image we're about to transfer therefore 
        // represents the light collected between tInt and the SH pulse we
        // just did.  The image covers a time range rather than a single 
        // point in time, but we still have to give it a single timestamp. 
        // Use the midpoint of the integration period.
        uint32_t tmid = (tNewInt + tInt) >> 1;
        if (pixDMA)
            t2 = tmid;
        else
            t1 = tmid;

        // Record the start time of the currently active integration period
        tInt = tNewInt;
    }
    
    // End of transfer notification.  This runs as an interrupt handler when
    // the DMA transfer completes.
    void transferDone()
    {
        // stop the ADC triggering
        os.stop();

        // add this sample to the timing statistics (for diagnostics and
        // performance measurement)
        uint32_t now = t.read_us();
        uint32_t dt = dtPixXfer = static_cast<uint32_t>(now - tXfer);
        totalXferTime += dt;
        nRuns += 1;
        
        // collect debug statistics
        if (dt < minXferTime) minXferTime = dt;
        if (dt > maxXferTime) maxXferTime = dt;

        // figure how long we've been integrating so far on this cycle 
        uint32_t dtInt = now - tInt;
        
        // Figure the time to the start of the next transfer.  Wait for the
        // remainder of the current integration period if we haven't yet
        // reached the requested minimum, otherwise just start almost
        // immediately.  (Not *actually* immediately: we don't want to start 
        // the new transfer within this interrupt handler, because the DMA
        // IRQ doesn't reliably clear if we start a new transfer immediately.)
        uint32_t dtNext = dtInt < tIntMin ? tIntMin - dtInt : 1;
        
        // Schedule the next transfer
        integrationTimeout.attach_us(this, &TCD1103::startTransfer, dtNext);
    }

    // Generate an SH/ICG pulse.  This transfers the pixel data from the live
    // sensor photoreceptors into the sensor's internal shift register, clears
    // the live pixels, and starts a new integration cycle.
    //
    // If start_dma_xfer is true, we'll start the DMA transfer for the ADC
    // pixel data.  We handle this here because the sensor starts clocking
    // out pixels precisely at the end of the ICG pulse, so we have to be
    // be very careful about the timing.
    //
    // Returns the timestamp (relative to our image timer 't') of the end
    // of the SH pulse, which is the moment the new integration cycle starts.
    //
    // Note that we send these pulses synchronously - that is, this routine
    // blocks until the pulses have been sent.  The overall sequence takes 
    // about 2.5us to 3us, so it's not a significant interruption of the 
    // main loop.
    //
    uint32_t gen_SH_ICG_pulse(bool start_dma_xfer)
    {
        // Make sure the ADC is stopped
        os.stop();

        // If desired, prepare to start the DMA transfer for the ADC data.
        // (Set up a dummy location to write in lieu of the DMA register if
        // DMA initiation isn't required, so that we don't have to take the
        // time for a conditional when we're ready to start the DMA transfer.
        // The timing there will be extremely tight, and we can't afford the
        // extra instructions to test a condition.)
        uint8_t dma_chcfg_dummy = 0;
        volatile uint8_t *dma_chcfg = start_dma_xfer ? os_dma.prepare(nPixSensor, true) : &dma_chcfg_dummy;
        
        // The basic idea is to take ICG low, and while holding ICG low,
        // pulse SH.  The coincidence of the two pulses transfers the charge
        // from the live pixels into the shift register, which effectively
        // discharges the live pixels and thereby starts a new integration
        // cycle.
        //
        // The timing of the pulse sequence is rather tightly constrained 
        // per the data sheet, so we have to take some care in executing it:
        //
        //   ICG ->  LOW
        //   100-1000 ns delay   (*)
        //   SH -> HIGH
        //   >1000ns delay
        //   SH -> LOW
        //   >1000ns delay
        //   ICG -> high         (**)
        //
        // There are two steps here that are tricky:
        //
        // (*) is a narrow window that we can't achieve with an mbed 
        // microsecond timer.  Instead, we'll do a couple of extra writes 
        // to the ICG register, which take about 60ns each.
        //
        // (**) has the rather severe constraint that the transition must 
        // occur AND complete while the master clock is high.  Other people 
        // working with similar Toshiba chips in MCU projects have suggested
        // that this constraint can safely be ignored, so maybe the data
        // sheet's insistence about it is obsolete advice from past Toshiba
        // sensors that the doc writers carried forward by copy-and-paste.
        // Toshiba has been making these sorts of chips for a very long time,
        // and the data sheets for many of them are obvious copy-and-paste
        // jobs.  But let's take the data sheet at its word and assume that 
        // this is important for proper operation.  Our best hope of 
        // satisfying this constraint is to synchronize the start of the
        // ICG->high transition with the start of a TPM cycle on the master
        // clock.  That guarantees that the ICG transition starts when the
        // clock signal is high (as each TPM cycle starts out high), and
        // gives us the longest possible runway for the transition to
        // complete while the clock is still high, as we get the full
        // length of the high part of the cycle to work with.  To quantify,
        // it gives us about 600ns.  The register write takes about 60ns, 
        // and waitEndCycle() adds several instructions of overhead, perhaps
        // 200ns, so we get around 300ns for the transition to finish.  That
        // should be a gracious plenty assuming that the hardware is set up 
        // with an inverter to buffer the clock signals.  The inverter should
        // be able to pull up the 35pF on ICG in a "typical" 30ns (rise time
        // plus propagation delay, per the 74HC04 data sheet) and max 150ns.
        // This seems to be one place where the inverter might really be
        // necessary to meet the timing requirements, as the KL25Z GPIO
        // might need more like 2us to pull that load up.
        //
        // There's an additional constraint on the timing at the end of the
        // ICG pulse.  The sensor starts clocking out pixels on the rising
        // edge of the ICG pulse.  So we need the ICG pulse end to align
        // with the start of an ADC cycle.  If we get that wrong, all of our
        // ADC samples will be off by half a clock, so every sample will be
        // the average of two adjacent pixels instead of one pixel.  That
        // would have the effect of shifting the image by half a pixel,
        // which could make our edge detection jitter by one pixel from one
        // frame to the next.  So we definitely want to avoid this.
        //
        // The end of the SH pulse triggers the start of a new integration 
        // cycle, so note the time of that pulse for image timestamping 
        // purposes.  That will be the start time of the NEXT image we 
        // transfer after we shift out the current sensor pixels, which 
        // represent the pixels from the last time we pulsed SH.
        //
        icg = logicLow;
        icg = logicLow;  // for timing, adds about 150ns > min 100ns

        sh = logicHigh;  // take SH high
        
        wait_us(1);      // >1000ns delay
        sh = logicHigh;  // a little more padding to be sure we're over the minimum
        
        sh = logicLow;   // take SH low
        
        uint32_t t_sh = t.read_us();  // this is the start time of the NEXT integration
        
        wait_us(3);      // >1000ns delay, 5000ns typical; 3us should get us most
                         // of the way there, considering that we have some more
                         // work to do before we end the ICG pulse
        
        // Now the tricky part!  We have to end the ICG pulse (take ICG high)
        // at the start of a master clock cycle, AND at the start of an ADC 
        // sampling cycle.  The sensor will start clocking out pixels the
        // instance ICG goes high, so we have to align our ADC cycle so that
        // we start a sample at almost exactly the same time we take ICG
        // high.
        //
        // Now, every ADC sampling cycle always starts at a rising edge of 
        // the master clock, since the master clock is the ADC trigger.  BUT,
        // the converse is NOT true: every rising edge of the master clock
        // is NOT an ADC sample start.  Recall that we've contrived the timing
        // so that every OTHER master clock rising edge starts an ADC sample.
        // 
        // So how do we detect which part of the clock cycle we're in?  We
        // could conceivably use the COCO bit in the ADC status register to
        // detect the little window between the end of one sample and the
        // start of the next.  Unfortunately, this doesn't work: the COCO
        // bit is never actually set for the duration of even a single CPU
        // instruction in our setup, no matter how loose we make the timing
        // between the ADC and the fM cycle.  I think the reason is the DMA
        // setup: the COCO bit triggers the DMA, and the DMA controller
        // reads the ADC result register (the DMA source in our setup),
        // which has the side effect of clearing COCO.  I've experimented
        // with this using different timing parameters, and the result is
        // always the same: the CPU *never* sees the COCO bit set.  The DMA
        // trigger timing is evidently deterministic such that the DMA unit
        // invariably gets its shot at reading ADC0->R before the CPU does.
        //
        // The COCO approach would be a little iffy anyway, since we want the
        // ADC idle time to be as short as possible, which wouldn't give us
        // much time to do all we have to do in the COCO period, even if
        // there were one.  What we can do instead is seize control of the
        // ADC cycle timing: rather than trying to detect when the cycle
        // ends, we can specify when it begins.  We can do this by canceling
        // the TPM->ADC trigger and aborting any conversion in progress, then
        // reprogramming the TPM->ADC trigger at our leisure.  What we *can*
        // detect reliably is the start of a TPM cycle.  So here's our
        // strategy:
        //
        //   - Turn off the TPM->ADC trigger and abort the current conversion
        //   - Wait until a new TPM cycle starts
        //   - Reset the TPM->ADC trigger.  The first new conversion will
        //     start on the next TPM cycle, so we have the remainder of
        //     the current TPM cycle to make this happen (about 1us, enough
        //     for 16 CPU instructions - plenty for this step)
        //   - Wait for the new TPM cycle
        //   - End the ICG pulse
        //
        
        // Enable the DMA controller for the new transfer from the ADC.
        // The sensor will start clocking out new samples at the ICG rising
        // edge, so the next ADC sample to complete will represent the first
        // pixel in the new frame.  So we need the DMA ready to go at the
        // very next sample.  Recall that the DMA is triggered by ADC
        // completion, and ADC is stopped right now, so enabling the DMA 
        // won't have any immediate effect - it just spools it up so that
        // it's ready to move samples as soon as we resume the ADC.
        *dma_chcfg |= DMAMUX_CHCFG_ENBL_MASK;
        
        // wait for the start of a new master clock cycle
        fm.waitEndCycle();
        
        // Wait one more cycle to be sure the DMA is ready.  Empirically,
        // this extra wait is actually required; evidently DMA startup has
        // some non-deterministic timing element or perhaps an asynchronous
        // external dependency.  In any case, *without* this extra wait,
        // the DMA transfer sporadically (about 20% probability) misses the
        // very first pixel that the sensor clocks out, so the entire image
        // is shifted "left" by one pixel.  That makes the position sensing
        // jitter by a pixel from one frame to the next according to whether
        // or not we had that one-pixel delay in the DMA startup.  Happily,
        // padding the timing by an fM cycle seems to make the DMA startup
        // perfectly reliable.
        fm.waitEndCycle();
        
        // Okay, a master clock cycle just started, so we have about 1us 
        // (about 16 CPU instructions) before the next one begins.  Resume 
        // ADC sampling.  The first new sample will start with the next
        // TPM cycle 1us from now.  This step itself takes about 3 machine
        // instructions for 180ns, so we have about 820ns left to go.
        os.resume();
        
        // Eerything is queued up!  We just have to fire the starting gun
        // on the sensor at the right moment.  And that right moment is the 
        // start of the next TPM cycle.  Wait for it...
        fm.waitEndCycle();
        
        // And go!
        icg = logicHigh;
        
        // note the start time of the transfer
        tXfer = t.read_us();
        
        // return the timestamp of the end of the SH pulse - this is the start
        // of the new integration period that we just initiated
        return t_sh;
    }

    // master clock
    NewPwmOut fm;
    
    // analog input for reading the pixel voltage level
    AltAnalogIn_8bit os;
    
    // Integration Clear Gate output
    DigitalOut icg;
    
    // Shift Gate output
    DigitalOut sh;
    
    // DMA channel for the analog input
    SimpleDMA os_dma;
    
    // Master clock period, in seconds, calculated based on the ADC timing
    float masterClockPeriod;
    
    // Number of pixels.  The TCD1103 has 1500 image pixels, plus 32 dummy
    // pixels at the front end (before the first image pixel) and another 14
    // dummy pixels at the back end.  The sensor always transfers the full
    // file on each read cycle, including the dummies, so we have to make
    // room for the dummy pixels during each read.
    static const int nPixSensor = 1546;
    
    // Figure the number of pixels to allocate per pixel buffer.  Round
    // up to the next 4-byte boundary, so that the buffers are both DWORD-
    // aligned.  (This allows using DWORD pointers into the buffer to 
    // operate on buffer pixels four at a time, such as in the negative 
    // image inversion code in the generic PlungerSensorImage base class.)
    static const int nPixAlo = (nPixSensor + 3) & ~3;
    
    // pixel buffers - we keep two buffers so that we can transfer the
    // current sensor data into one buffer via DMA while we concurrently
    // process the last buffer
    uint8_t *pix1;            // pixel array 1
    uint8_t *pix2;            // pixel array 2
    
    // Timestamps of pix1 and pix2 arrays, in microseconds, in terms of the 
    // sample timer (this->t).
    uint32_t t1;
    uint32_t t2;
    
    // DMA target buffer.  This is the buffer for the next DMA transfer.
    // 0 means pix1, 1 means pix2.  The other buffer contains the stable 
    // data from the last transfer.
    uint8_t pixDMA;
    
    // Stable buffer ownership.  At any given time, the DMA subsystem owns
    // the buffer specified by pixDMA.  The other buffer - the "stable" buffer,
    // which contains the most recent completed frame, can be owned by EITHER
    // the client or by the DMA subsystem.  Each time a DMA transfer completes,
    // the DMA subsystem looks at the stable buffer owner flag to determine 
    // what to do:
    //
    // - If the DMA subsystem owns the stable buffer, it swaps buffers.  This
    //   makes the newly completed DMA buffer the new stable buffer, and makes
    //   the old stable buffer the new DMA buffer.  At this time, the DMA 
    //   subsystem also changes the stable buffer ownership to CLIENT.
    //
    // - If the CLIENT owns the stable buffer, the DMA subsystem can't swap
    //   buffers, because the client is still using the stable buffer.  It
    //   simply leaves things as they are.
    //
    // In either case, the DMA system starts a new transfer at this point.
    //
    // The client, meanwhile, is free to access the stable buffer when it has
    // ownership.  If the client *doesn't* have ownership, it must wait for
    // the ownership to be transferred, which can only be done by the DMA
    // subsystem on completing a transfer.
    //
    // When the client is done with the stable buffer, it transfers ownership
    // back to the DMA subsystem.
    //
    // Transfers of ownership from DMA to CLIENT are done only by DMA.
    // Transfers from CLIENT to DMA are done only by CLIENT.  So whoever has
    // ownership now is responsible for transferring ownership.
    //
    volatile bool clientOwnsStablePix;
    
    // Minimum requested integration time, in microseconds
    uint32_t tIntMin;
    
    // Timeout for generating an interrupt at the end of the integration period
    Timeout integrationTimeout;
        
    // timing statistics
    Timer t;                  // sample timer
    uint32_t tInt;            // start time (us) of current integration period
    uint32_t tXfer;           // start time (us) of current pixel transfer
    uint32_t dtPixXfer;       // pixel transfer time of last frame
    uint64_t totalXferTime;   // total time consumed by all reads so far
    uint32_t nRuns;           // number of runs so far
    
    // debugging - min/max transfer time statistics
    uint32_t minXferTime;
    uint32_t maxXferTime;
};