An input/output controller for virtual pinball machines, with plunger position tracking, accelerometer-based nudge sensing, button input encoding, and feedback device control.

Dependencies:   USBDevice mbed FastAnalogIn FastIO FastPWM SimpleDMA

/media/uploads/mjr/pinscape_no_background_small_L7Miwr6.jpg

The Pinscape Controller is a special-purpose software project that I wrote for my virtual pinball machine.

New version: V2 is now available! The information below is for version 1, which will continue to be available for people who prefer the original setup.

What exactly is a virtual pinball machine? It's basically a video-game pinball emulator built to look like a real pinball machine. (The picture at right is the one I built.) You start with a standard pinball cabinet, either built from scratch or salvaged from a real machine. Inside, you install a PC motherboard to run the software, and install TVs in place of the playfield and backglass. Several Windows pinball programs can take advantage of this setup, including the open-source project Visual Pinball, which has hundreds of tables available. Building one of these makes a great DIY project, and it's a good way to add to your skills at woodworking, computers, and electronics. Check out the Cabinet Builders' Forum on vpforums.org for lots of examples and advice.

This controller project is a key piece in my setup that helps integrate the video game into the pinball cabinet. It handles several input/output tasks that are unique to virtual pinball machines. First, it lets you connect a mechanical plunger to the software, so you can launch the ball like on a real machine. Second, it sends "nudge" data to the software, based on readings from an accelerometer. This lets you interact with the game physically, which makes the playing experience more realistic and immersive. Third, the software can handle button input (for wiring flipper buttons and other cabinet buttons), and fourth, it can control output devices (for tactile feedback, button lights, flashers, and other special effects).

Documentation

The Hardware Build Guide (PDF) has detailed instructions on how to set up a Pinscape Controller for your own virtual pinball cabinet.

Update notes

December 2015 version: This version fully supports the new Expansion Board project, but it'll also run without it. The default configuration settings haven't changed, so existing setups should continue to work as before.

August 2015 version: Be sure to get the latest version of the Config Tool for windows if you're upgrading from an older version of the firmware. This update adds support for TSL1412R sensors (a version of the 1410 sensor with a slightly larger pixel array), and a config option to set the mounting orientation of the board in the firmware rather than in VP (for better support for FP and other pinball programs that don't have VP's flexibility for setting the rotation).

Feb/March 2015 software versions: If you have a CCD plunger that you've been using with the older versions, and the plunger stops working (or doesn't work as well) after you update to the latest version, you might need to increase the brightness of your light source slightly. Check the CCD exposure with the Windows config tool to see if it looks too dark. The new software reads the CCD much more quickly than the old versions did. This makes the "shutter speed" faster, which might require a little more light to get the same readings. The CCD is actually really tolerant of varying light levels, so you probably won't have to change anything for the update - I didn't. But if you do have any trouble, have a look at the exposure meter and try a slightly brighter light source if the exposure looks too dark.

Downloads

  • Config tool for Windows (.exe and C# source): this is a Windows program that lets you view the raw pixel data from the CCD sensor, trigger plunger calibration mode, and configure some of the software options on the controller.
  • 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 9.9.1 and VP 10 releases, so you don't need my custom builds if you're using 9.9.1 or 10 or later. I don't think there's any reason to use my 9.9 instead of the official 9.9.1, but I'm leaving it here just in case. In the official VP releases, look for the checkbox "Enable Nudge Filter" in the Keys preferences dialog. (There's no checkbox in my custom builds, though; the filter is simply always on in those.)
  • Output circuit shopping list: This is a saved shopping cart at mouser.com with the parts needed for each output driver, if you want to use the LedWiz emulator feature. Note that quantities in the cart are for one output channel, so multiply everything by the number of channels you plan to use, except that you only need one of the ULN2803 transistor array chips for each eight output circuits.
  • 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.

Features

  • Plunger position sensing, using a TAOS TSL 1410R CCD linear array sensor. This sensor is a 1280 x 1 pixel array at 400 dpi, which makes it about 3" long - almost exactly the travel distance of a standard pinball plunger. The idea is that you install the sensor just above (within a few mm of) the shooter rod on the inside of the cabinet, with the CCD window facing down, aligned with and centered on the long axis of the shooter rod, and positioned so that the rest position of the tip is about 1/2" from one end of the window. As you pull back the plunger, the tip will travel down the length of the window, and the maximum retraction point will put the tip just about at the far end of the window. Put a light source below, facing the sensor - I'm using two typical 20 mA blue LEDs about 8" away (near the floor of the cabinet) with good results. The principle of operation is that the shooter rod casts a shadow on the CCD, so pixels behind the rod will register lower brightness than pixels that aren't in the shadow. We scan down the length of the sensor for the edge between darker and brighter, and this tells us how far back the rod has been pulled. We can read the CCD at about 25-30 ms intervals, so we can get rapid updates. We pass the readings reports to VP via our USB joystick reports.

    The hardware build guide includes schematics showing how to wire the CCD to the KL25Z. It's pretty straightforward - five wires between the two devices, no external components needed. Two GPIO ports are used as outputs to send signals to the device and one is used as an ADC in to read the pixel brightness inputs. The config tool has a feature that lets you display the raw pixel readings across the array, so you can test that the CCD is working and adjust the light source to get the right exposure level.

    Alternatively, you can use a slide potentiometer as the plunger sensor. This is a cheaper and somewhat simpler option that seems to work quite nicely, as you can see in Lemming77's video of this setup in action. This option is also explained more fully in the build guide.
  • Nudge sensing via the KL25Z's on-board accelerometer. Mounting the board in your cabinet makes it feel the same accelerations the cabinet experiences when you nudge it. Visual Pinball already knows how to interpret accelerometer input as nudging, so we simply feed the acceleration readings to VP via the joystick interface.
  • Cabinet button wiring. Up to 24 pushbuttons and switches can be wired to the controller for input controls (for example, flipper buttons, the Start button, the tilt bob, coin slot switches, and service door buttons). These appear to Windows as joystick buttons. VP can map joystick buttons to pinball inputs via its keyboard preferences dialog. (You can raise the 24-button limit by editing the source code, but since all of the GPIO pins are allocated, you'll have to reassign pins currently used for other functions.)
  • LedWiz emulation (limited). In addition to emulating a joystick, the device emulates the LedWiz USB interface, so controllers on the PC side such as DirectOutput Framework can recognize it and send it commands to control lights, solenoids, and other feedback devices. 22 GPIO ports are assigned by default as feedback device outputs. This feature has some limitations. The big one is that the KL25Z hardware only has 10 PWM channels, which isn't enough for a fully decked-out cabinet. You also need to build some external power driver circuitry to use this feature, because of the paltry 4mA output capacity of the KL25Z GPIO ports. The build guide includes instructions for a simple and robust output circuit, including part numbers for the exact components you need. It's not hard if you know your way around a soldering iron, but just be aware that it'll take a little work.

Warning: This is not replacement software for the VirtuaPin plunger kit. If you bought the VirtuaPin kit, please don't try to install this software. The VP kit happens to use the same microcontroller board, but the rest of its hardware is incompatible. The VP kit uses a different type of sensor for its plunger and has completely different button wiring, so the Pinscape software won't work properly with it.

main.cpp

Committer:
mjr
Date:
2014-07-24
Revision:
4:02c7cd7b2183
Parent:
3:3514575d4f86
Child:
5:a70c0bce770d

File content as of revision 4:02c7cd7b2183:

#include "mbed.h"
#include "USBJoystick.h"
#include "MMA8451Q.h"
#include "tsl1410r.h"
#include "FreescaleIAP.h"
#include "crc32.h"

// customization of the joystick class to expose connect/suspend status
class MyUSBJoystick: public USBJoystick
{
public:
    MyUSBJoystick(uint16_t vendor_id, uint16_t product_id, uint16_t product_release) 
        : USBJoystick(vendor_id, product_id, product_release, false)
    {
        suspended_ = false;
    }
    
    int isConnected() { return configured(); }
    int isSuspended() const { return suspended_; }
    
protected:
    virtual void suspendStateChanged(unsigned int suspended)
        { suspended_ = suspended; }

    int suspended_; 
};

// On-board RGB LED elements - we use these for diagnostic displays.
DigitalOut ledR(LED1), ledG(LED2), ledB(LED3);

// calibration button - switch input and LED output
DigitalIn calBtn(PTE29);
DigitalOut calBtnLed(PTE23);

static int pbaIdx = 0;

// on/off state for each LedWiz output
static uint8_t wizOn[32];

// profile (brightness/blink) state for each LedWiz output
static uint8_t wizVal[32] = {
    0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0
};

static float wizState(int idx)
{
    if (wizOn[idx]) {
        // on - map profile brightness state to PWM level
        uint8_t val = wizVal[idx];
        if (val >= 1 && val <= 48)
            return 1.0 - val/48.0;
        else if (val >= 129 && val <= 132)
            return 0.0;
        else
            return 1.0;
    }
    else {
        // off
        return 1.0;
    }
}

static void updateWizOuts()
{
    ledR = wizState(0);
    ledG = wizState(1);
    ledB = wizState(2);
}

struct AccPrv
{
    AccPrv() : x(0), y(0) { }
    float x;
    float y;
    
    double dist(AccPrv &b)
    {
        float dx = x - b.x, dy = y - b.y;
        return sqrt(dx*dx + dy*dy);
    }
};

// Non-volatile memory structure.  We store persistent a small
// amount of persistent data in flash memory to retain calibration
// data between sessions.
struct NVM
{
    // checksum - we use this to determine if the flash record
    // has been initialized
    uint32_t checksum;

    // signature value
    static const uint32_t SIGNATURE = 0x4D4A522A;
    static const uint16_t VERSION = 0x0002;
    
    // stored data (excluding the checksum)
    struct
    {
        // signature and version - further verification that we have valid 
        // initialized data
        uint32_t sig;
        uint16_t vsn;
        
        // direction - 0 means unknown, 1 means bright end is pixel 0, 2 means reversed
        uint8_t dir;

        // plunger calibration min and max
        int plungerMin;
        int plungerMax;
    } d;
};

// Accelerometer handler
const int MMA8451_I2C_ADDRESS = (0x1d<<1);
class Accel
{
public:
    Accel(PinName sda, PinName scl, int i2cAddr, PinName irqPin)
        : mma_(sda, scl, i2cAddr), intIn_(irqPin)
    {
        // set the initial ball velocity to zero
        vx_ = vy_ = 0;
        
        // set the initial raw acceleration reading to zero
        xRaw_ = yRaw_ = 0;

        // enable the interrupt
        mma_.setInterruptMode(irqPin == PTA14 ? 1 : 2);
        
        // set up the interrupt handler
        intIn_.rise(this, &Accel::isr);
        
        // read the current registers to clear the data ready flag
        float z;
        mma_.getAccXYZ(xRaw_, yRaw_, z);

        // start our timers
        tGet_.start();
        tInt_.start();
    }
    
    void get(float &x, float &y, float &rx, float &ry) 
    {
         // disable interrupts while manipulating the shared data
         __disable_irq();
         
         // read the shared data and store locally for calculations
         float vx = vx_, vy = vy_, xRaw = xRaw_, yRaw = yRaw_;

         // reset the velocity
         vx_ = vy_ = 0;
         
         // get the time since the last get() sample
         float dt = tGet_.read_us()/1.0e6;
         tGet_.reset();
         
         // done manipulating the shared data
         __enable_irq();
         
         // calculate the acceleration since the last get(): a = dv/dt
         x = vx/dt;
         y = vy/dt;         
         
         // return the raw accelerometer data in rx,ry
         rx = xRaw;
         ry = yRaw;
     }    
    
private:
    // interrupt handler
    void isr()
    {
        // Read the axes.  Note that we have to read all three axes
        // (even though we only really use x and y) in order to clear
        // the "data ready" status bit in the accelerometer.  The
        // interrupt only occurs when the "ready" bit transitions from
        // off to on, so we have to make sure it's off.
        float z;
        mma_.getAccXYZ(xRaw_, yRaw_, z);
        
        // calculate the time since the last interrupt
        float dt = tInt_.read_us()/1.0e6;
        tInt_.reset();
        
        // Accelerate the model ball: v = a*dt.  Assume that the raw
        // data from the accelerometer reflects the average physical
        // acceleration over the interval since the last sample.
        vx_ += xRaw_ * dt;
        vy_ += yRaw_ * dt;
    }
    
    // current modeled ball velocity
    float vx_, vy_;
    
    // last raw axis readings
    float xRaw_, yRaw_;
    
    // underlying accelerometer object
    MMA8451Q mma_;
    
    // interrupt router
    InterruptIn intIn_;
    
    // timer for measuring time between get() samples
    Timer tGet_;
    
    // timer for measuring time between interrupts
    Timer tInt_;
};

int main(void)
{
    // turn off our on-board indicator LED
    ledR = 1;
    ledG = 1;
    ledB = 1;
    
    // set up a flash memory controller
    FreescaleIAP iap;
    
    // use the last sector of flash for our non-volatile memory structure
    int flash_addr = (iap.flash_size() - SECTOR_SIZE);
    NVM *flash = (NVM *)flash_addr;
    NVM cfg;
    
    // check for valid flash
    bool flash_valid = (flash->d.sig == flash->SIGNATURE 
                        && flash->d.vsn == flash->VERSION
                        && flash->checksum == CRC32(&flash->d, sizeof(flash->d)));
                      
    // Number of pixels we read from the sensor on each frame.  This can be
    // less than the physical pixel count if desired; we'll read every nth
    // piexl if so.  E.g., with a 1280-pixel physical sensor, if npix is 320,
    // we'll read every 4th pixel.  VP doesn't seem to have very high
    // resolution internally for the plunger, so it's probably not necessary
    // to use the full resolution of the sensor - about 160 pixels seems
    // perfectly adequate.  We can read the sensor faster (and thus provide
    // a higher refresh rate) if we read fewer pixels in each frame.
    const int npix = 160;

    // if the flash is valid, load it; otherwise initialize to defaults
    if (flash_valid) {
        memcpy(&cfg, flash, sizeof(cfg));
        printf("Flash restored: plunger min=%d, max=%d\r\n", 
            cfg.d.plungerMin, cfg.d.plungerMax);
    }
    else {
        printf("Factory reset\r\n");
        cfg.d.sig = cfg.SIGNATURE;
        cfg.d.vsn = cfg.VERSION;
        cfg.d.plungerMin = 0;
        cfg.d.plungerMax = npix;
    }
    
    // plunger calibration button debounce timer
    Timer calBtnTimer;
    calBtnTimer.start();
    int calBtnDownTime = 0;
    int calBtnLit = false;
    
    // Calibration button state:
    //  0 = not pushed
    //  1 = pushed, not yet debounced
    //  2 = pushed, debounced, waiting for hold time
    //  3 = pushed, hold time completed - in calibration mode
    int calBtnState = 0;
    
    // set up a timer for our heartbeat indicator
    Timer hbTimer;
    hbTimer.start();
    int t0Hb = hbTimer.read_ms();
    int hb = 0;
    
    // set a timer for accelerometer auto-centering
    Timer acTimer;
    acTimer.start();
    int t0ac = acTimer.read_ms();
    
    // Create the joystick USB client
    MyUSBJoystick js(0xFAFA, 0x00F7, 0x0003);

    // create the accelerometer object
    Accel accel(PTE25, PTE24, MMA8451_I2C_ADDRESS, PTA15);
    
    // create the CCD array object
    TSL1410R ccd(PTE20, PTE21, PTB0);
    
    // recent accelerometer readings, for auto centering
    int iAccPrv = 0, nAccPrv = 0;
    const int maxAccPrv = 5;
    AccPrv accPrv[maxAccPrv];

    // last accelerometer report, in mouse coordinates
    int x = 127, y = 127, z = 0;

    // raw accelerator centerpoint, on the unit interval (-1.0 .. +1.0)
    float xCenter = 0.0, yCenter = 0.0;    
    
    // start the first CCD integration cycle
    ccd.clear();

    // we're all set up - now just loop, processing sensor reports and 
    // host requests
    for (;;)
    {
        // Look for an incoming report.  Continue processing input as
        // long as there's anything pending - this ensures that we
        // handle input in as timely a fashion as possible by deferring
        // output tasks as long as there's input to process.
        HID_REPORT report;
        while (js.readNB(&report) && report.length == 8)
        {
            uint8_t *data = report.data;
            if (data[0] == 64) 
            {
                // LWZ-SBA - first four bytes are bit-packed on/off flags
                // for the outputs; 5th byte is the pulse speed (0-7)
                //printf("LWZ-SBA %02x %02x %02x %02x ; %02x\r\n",
                //       data[1], data[2], data[3], data[4], data[5]);

                // update all on/off states
                for (int i = 0, bit = 1, ri = 1 ; i < 32 ; ++i, bit <<= 1)
                {
                    if (bit == 0x100) {
                        bit = 1;
                        ++ri;
                    }
                    wizOn[i] = ((data[ri] & bit) != 0);
                }
    
                // update the physical outputs
                updateWizOuts();
                
                // reset the PBA counter
                pbaIdx = 0;
            }
            else 
            {
                // LWZ-PBA - full state dump; each byte is one output
                // in the current bank.  pbaIdx keeps track of the bank;
                // this is incremented implicitly by each PBA message.
                //printf("LWZ-PBA[%d] %02x %02x %02x %02x %02x %02x %02x %02x\r\n",
                //       pbaIdx, data[0], data[1], data[2], data[3], data[4], data[5], data[6], data[7]);

                // update all output profile settings
                for (int i = 0 ; i < 8 ; ++i)
                    wizVal[pbaIdx + i] = data[i];

                // update the physical LED state if this is the last bank                    
                if (pbaIdx == 24)
                    updateWizOuts();

                // advance to the next bank
                pbaIdx = (pbaIdx + 8) & 31;
            }
        }
       
        // check for plunger calibration
        if (!calBtn)
        {
            // check the state
            switch (calBtnState)
            {
            case 0: 
                // button not yet pushed - start debouncing
                calBtnTimer.reset();
                calBtnDownTime = calBtnTimer.read_ms();
                calBtnState = 1;
                break;
                
            case 1:
                // pushed, not yet debounced - if the debounce time has
                // passed, start the hold period
                if (calBtnTimer.read_ms() - calBtnDownTime > 50)
                    calBtnState = 2;
                break;
                
            case 2:
                // in the hold period - if the button has been held down
                // for the entire hold period, move to calibration mode
                if (calBtnTimer.read_ms() - calBtnDownTime > 2050)
                {
                    // enter calibration mode
                    calBtnState = 3;
                    
                    // reset the calibration limits
                    cfg.d.plungerMax = 0;
                    cfg.d.plungerMin = npix;
                }
                break;
                
            case 3:
                // Already in calibration mode - pushing the button in this
                // state doesn't change the current state, but we won't leave
                // this state as long as it's held down.  We can simply do
                // nothing here.
                break;
            }
        }
        else
        {
            // Button released.  If we're in calibration mode, and
            // the calibration time has elapsed, end the calibration
            // and save the results to flash.
            //
            // Otherwise, return to the base state without saving anything.
            // If the button is released before we make it to calibration
            // mode, it simply cancels the attempt.
            if (calBtnState == 3
                && calBtnTimer.read_ms() - calBtnDownTime > 17500)
            {
                // exit calibration mode
                calBtnState = 0;
                
                // Save the current configuration state to flash, so that it
                // will be preserved through power off.  Update the checksum
                // first so that we recognize the flash record as valid.
                cfg.checksum = CRC32(&cfg.d, sizeof(cfg.d));
                iap.erase_sector(flash_addr);
                iap.program_flash(flash_addr, &cfg, sizeof(cfg));
                
                // the flash state is now valid
                flash_valid = true;
            }
            else if (calBtnState != 3)
            {
                // didn't make it to calibration mode - cancel the operation
                calBtnState = 0;
            }
        }       
        
        // light/flash the calibration button light, if applicable
        int newCalBtnLit = calBtnLit;
        switch (calBtnState)
        {
        case 2:
            // in the hold period - flash the light
            newCalBtnLit = (((calBtnTimer.read_ms() - calBtnDownTime)/250) & 1);
            break;
            
        case 3:
            // calibration mode - show steady on
            newCalBtnLit = true;
            break;
            
        default:
            // not calibrating/holding - show steady off
            newCalBtnLit = false;
            break;
        }
        
        // light or flash the external calibration button LED, and 
        // do the same with the on-board blue LED
        if (calBtnLit != newCalBtnLit)
        {
            calBtnLit = newCalBtnLit;
            if (calBtnLit) {
                calBtnLed = 1;
                ledR = 1;
                ledG = 1;
                ledB = 1;
            }
            else {
                calBtnLed = 0;
                ledR = 1;
                ledG = 1;
                ledB = 0;
            }
        }
        
        // read the plunger sensor
        int znew = z;
        uint16_t pix[npix];
        ccd.read(pix, npix);

        // get the average brightness at each end of the sensor
        long avg1 = (long(pix[0]) + long(pix[1]) + long(pix[2]) + long(pix[3]) + long(pix[4]))/5;
        long avg2 = (long(pix[npix-1]) + long(pix[npix-2]) + long(pix[npix-3]) + long(pix[npix-4]) + long(pix[npix-5]))/5;
        
        // figure the midpoint in the brightness; multiply by 3 so that we can
        // compare sums of three pixels at a time to smooth out noise
        long midpt = (avg1 + avg2)/2 * 3;
        
        // Work from the bright end to the dark end.  VP interprets the
        // Z axis value as the amount the plunger is pulled: the minimum
        // is the rest position, the maximum is fully pulled.  So we 
        // essentially want to report how much of the sensor is lit,
        // since this increases as the plunger is pulled back.
        int si = 1, di = 1;
        if (avg1 < avg2)
            si = npix - 2, di = -1;

        // scan for the midpoint     
        uint16_t *pixp = pix + si;           
        for (int n = 1 ; n < npix - 1 ; ++n, pixp += di)
        {
            // if we've crossed the midpoint, report this position
            if (long(pixp[-1]) + long(pixp[0]) + long(pixp[1]) < midpt)
            {
                // note the new position
                int pos = n;
                
                // if the bright end and dark end don't differ by enough, skip this
                // reading entirely - we must have an overexposed or underexposed frame
                if (labs(avg1 - avg2) < 0x3333)
                    break; 
                
                // Calibrate, or apply calibration, depending on the mode.
                // In either case, normalize to a 0-127 range.  VP appears to
                // ignore negative Z axis values.
                if (calBtnState == 3)
                {
                    // calibrating - note if we're expanding the calibration envelope
                    if (pos < cfg.d.plungerMin)
                        cfg.d.plungerMin = pos;   
                    if (pos > cfg.d.plungerMax)
                        cfg.d.plungerMax = pos;
                        
                    // normalize to the full physical range while calibrating
                    znew = int(float(pos)/npix * 127);
                }
                else
                {
                    // running normally - normalize to the calibration range
                    if (pos < cfg.d.plungerMin)
                        pos = cfg.d.plungerMin;
                    if (pos > cfg.d.plungerMax)
                        pos = cfg.d.plungerMax;
                    znew = int(float(pos - cfg.d.plungerMin)
                        / (cfg.d.plungerMax - cfg.d.plungerMin + 1) * 127);
                }
                
                // done
                break;
            }
        }
        
        // read the accelerometer
        float xa, ya, rxa, rya;
        accel.get(xa, ya, rxa, rya);
        
        // check for auto-centering every so often
        if (acTimer.read_ms() - t0ac > 1000) 
        {
            // add the sample to the history list
            accPrv[iAccPrv].x = xa;
            accPrv[iAccPrv].y = ya;
            
            // store the slot
            iAccPrv += 1;
            iAccPrv %= maxAccPrv;
            nAccPrv += 1;
            
            // If we have a full complement, check for stability.  The
            // raw accelerometer input is in the rnage -4096 to 4096, but
            // the class cover normalizes to a unit interval (-1.0 .. +1.0).
            const float accTol = .005;
            if (nAccPrv >= maxAccPrv
                && accPrv[0].dist(accPrv[1]) < accTol
                && accPrv[0].dist(accPrv[2]) < accTol
                && accPrv[0].dist(accPrv[3]) < accTol
                && accPrv[0].dist(accPrv[4]) < accTol)
            {
                // figure the new center
                xCenter = (accPrv[0].x + accPrv[1].x + accPrv[2].x + accPrv[3].x + accPrv[4].x)/5.0;
                yCenter = (accPrv[0].y + accPrv[1].y + accPrv[2].y + accPrv[3].y + accPrv[4].y)/5.0;
            }
            
            // reset the auto-center timer
            acTimer.reset();
            t0ac = acTimer.read_ms();
        }
        
        // adjust for our auto centering
        xa -= xCenter;
        ya -= yCenter;
        
        // confine to the unit interval
        if (xa < -1.0) xa = -1.0;
        if (xa > 1.0) xa = 1.0;
        if (ya < -1.0) ya = -1.0;
        if (ya > 1.0) ya = 1.0;

        // figure the new mouse report data
        int xnew = (int)(127 * xa);
        int ynew = (int)(127 * ya);

        // store the updated joystick coordinates
        x = xnew;
        y = ynew;
        z = znew;
        
        // if we're in USB suspend or disconnect mode, spin
        if (js.isSuspended() || !js.isConnected())
        {
            // go dark (turn off the indicator LEDs)
            ledG = 1;
            ledB = 1;
            ledR = 1;
            
            // wait until we're connected and come out of suspend mode
            for (uint32_t n = 0 ; js.isSuspended() || !js.isConnected() ; ++n)
            {
                // spin for a bit
                wait(1);
                
                // if we're suspended, do a brief red flash; otherwise do a long red flash
                if (js.isSuspended())
                {
                    // suspended - flash briefly ever few seconds
                    if (n % 3 == 0)
                    {
                        ledR = 0;
                        wait(0.05);
                        ledR = 1;
                    }
                }
                else
                {
                    // running, not connected - flash red
                    ledR = !ledR;
                }
            }
        }

        // Send the status report.  It doesn't really matter what
        // coordinate system we use, since Visual Pinball has config
        // options for rotations and axis reversals, but reversing y
        // at the device level seems to produce the most intuitive 
        // results for the Windows joystick control panel view, which
        // is an easy way to check that the device is working.
        js.update(x, -y, z, int(rxa*127), int(rya*127), 0);
        
        // show a heartbeat flash in blue every so often if not in 
        // calibration mode
        if (calBtnState < 2 && hbTimer.read_ms() - t0Hb > 1000) 
        {
            if (js.isSuspended())
            {
                // suspended - turn off the LEDs entirely
                ledR = 1;
                ledG = 1;
                ledB = 1;
            }
            else if (!js.isConnected())
            {
                // not connected - flash red
                hb = !hb;
                ledR = (hb ? 0 : 1);
                ledG = 1;
                ledB = 1;
            }
            else if (flash_valid)
            {
                // connected, NVM valid - flash blue/green
                hb = !hb;
                ledR = 1;
                ledG = (hb ? 0 : 1);
                ledB = (hb ? 1 : 0);
            }
            else
            {
                // connected, factory reset - flash yellow/green
                hb = !hb;
                ledR = (hb ? 0 : 1);
                ledG = 0;
                ledB = 1;
            }
            
            // reset the heartbeat timer
            hbTimer.reset();
            t0Hb = hbTimer.read_ms();
        }
    }
}