Advanced programming tips, tricks and hacks for Mac development in C/Objective-C and Cocoa.

An iOS tone generator (an introduction to AudioUnits)

In this post, I present a tiny iOS app that generates a continuous tone at a frequency determined by a slider. It's a small sample app intended to show the simplest way to send audio data you generate to the speaker.


I've previously written posts on playing audio from MP3/AAC files and streams. These posts use the AudioQueue API to play audio. The AudioQueue interface can take audio data — still in compressed formats — and play it on the device output. This decoding/decompressing is the key strength of the AudioQueue APIs and it is the only way to take advantage of the hardware decoding on iOS devices.

However, if you're generating your own audio (and it is therefore already decompressed, linear PCM) you don't need to use AudioQueue. You can still play audio in this format using AudioQueue but the reality is that you'll have more control and be able to do much more if you use the lowest level audio API (lowest in iOS, anyway): the AudioUnit.

Sample application: ToneGenerator

The sample application for this post is really simple: a slider to control frequency and a play button.


The tone is generated continuously while you change the frequency so you can play with the tone generator like a slide whistle for hours of neighbour-annoying fun.

You can download the ToneGenerator and the complete sample project used in this post here (25kb)

Audio Units

AudioUnits are lowest level of sound generation on iOS and the lowest hardware abstracted layer commonly used on the Mac. They generate raw audio samples when requested and place them into output buffers. That's their entire function.

An AudioUnit generates these samples in its render function. The render function is invoked to generate samples on a dedicated audio thread.

In an effort to keep the code as focussed as possible on the basics of AudioUnits, I will be showing you how to create a single AudioUnit and use it on its own to output sound. Ordinarily though, AudioUnits are used as part of an AUGraph which chains a few different units together for feeding in sound data, mixing, effects and other purposes. Perhaps I'll show more about that in a future post but for now, the aim is to keep it simple.

Generating your own audio samples

Generating audio samples means that we must calculate the values of an audio waveform at the required time locations (sample points).

In this sample application, we're going to use 32-bit floating point values for each sample (since that's the easiest). This is for convenience, it's not the best format for performance on iOS — that would be the canonical sample format of 8.24 signed, fixed-point samples (that's 32-bits per sample with 24 bits of that used as the fractional component). But convenience is more important than raw efficiency in this sample application so floats are good enough.

The values can vary between -1.0 and +1.0 but we're going to limit our samples to a smaller range than that to keep the volume to a reasonable level.

Our generated tone is a basic sine wave (a pure tone). The value at each sample point will then be determined by the following equation:

f(n) = a sin ( θ(n) )

where n is the index of the current sample, a is the amplitude and the current angle of the waveform, θ(n) is given by:

θ(n) = 2πƒ n / r

where ƒ is the frequency of the tone we want to generate and r is the sample rate of the audio we're generating.

Implementing this in our AudioUnit's render function, gives us a function that looks like this:

OSStatus RenderTone(
    void *inRefCon, 
    AudioUnitRenderActionFlags *ioActionFlags, 
    const AudioTimeStamp *inTimeStamp, 
    UInt32 inBusNumber, 
    UInt32 inNumberFrames, 
    AudioBufferList *ioData)

    // Fixed amplitude is good enough for our purposes
    const double amplitude = 0.25;

    // Get the tone parameters out of the view controller
    ToneGeneratorViewController *viewController =
        (ToneGeneratorViewController *)inRefCon;
    double theta = viewController->theta;
    double theta_increment =
        2.0 * M_PI * viewController->frequency / viewController->sampleRate;

    // This is a mono tone generator so we only need the first buffer
    const int channel = 0;
    Float32 *buffer = (Float32 *)ioData->mBuffers[channel].mData;
    // Generate the samples
    for (UInt32 frame = 0; frame < inNumberFrames; frame++) 
        buffer[frame] = sin(theta) * amplitude;
        theta += theta_increment;
        if (theta > 2.0 * M_PI)
            theta -= 2.0 * M_PI;
    // Store the updated theta back in the view controller
    viewController->theta = theta;

    return noErr;

Creating an AudioUnit for outputting audio

AudioUnit is a low-level API, so there are many settings you can configure — and you have to configure a lot of them to get playback to work at all. Fortunately, it's all pretty straightforward.

Here's the code to create and configure an output audio unit (toneUnit) for playing 32 bit, single channel, floating point, linear PCM sound that generates its audio using the function RenderTone.

// Configure the search parameters to find the default playback output unit
// (called the kAudioUnitSubType_RemoteIO on iOS but
// kAudioUnitSubType_DefaultOutput on Mac OS X)
AudioComponentDescription defaultOutputDescription;
defaultOutputDescription.componentType = kAudioUnitType_Output;
defaultOutputDescription.componentSubType = kAudioUnitSubType_RemoteIO;
defaultOutputDescription.componentManufacturer = kAudioUnitManufacturer_Apple;
defaultOutputDescription.componentFlags = 0;
defaultOutputDescription.componentFlagsMask = 0;

// Get the default playback output unit
AudioComponent defaultOutput = AudioComponentFindNext(NULL, &defaultOutputDescription);
NSAssert(defaultOutput, @"Can't find default output");

// Create a new unit based on this that we'll use for output
OSErr err = AudioComponentInstanceNew(defaultOutput, &toneUnit);
NSAssert1(toneUnit, @"Error creating unit: %ld", err);

// Set our tone rendering function on the unit
AURenderCallbackStruct input;
input.inputProc = RenderTone;
input.inputProcRefCon = self;
err = AudioUnitSetProperty(toneUnit, 
NSAssert1(err == noErr, @"Error setting callback: %ld", err);

// Set the format to 32 bit, single channel, floating point, linear PCM
const int four_bytes_per_float = 4;
const int eight_bits_per_byte = 8;
AudioStreamBasicDescription streamFormat;
streamFormat.mSampleRate = sampleRate;
streamFormat.mFormatID = kAudioFormatLinearPCM;
streamFormat.mFormatFlags =
    kAudioFormatFlagsNativeFloatPacked | kAudioFormatFlagIsNonInterleaved;
streamFormat.mBytesPerPacket = four_bytes_per_float;
streamFormat.mFramesPerPacket = 1;    
streamFormat.mBytesPerFrame = four_bytes_per_float;        
streamFormat.mChannelsPerFrame = 1;    
streamFormat.mBitsPerChannel = four_bytes_per_float * eight_bits_per_byte;
err = AudioUnitSetProperty (toneUnit,
NSAssert1(err == noErr, @"Error setting stream format: %ld", err);

Start it playing

Once you've got an output AudioUnit created, you need to initialize it using AudioUnitInitialize (which verifies that all the parameters are valid) and then you can start it running using AudioOutputUnitStart. Once an AudioUnit is initialized, you can't change further parameters, so if you need to change parameters again, you'll need to use AudioUnitUninitialize.

In the sample program, toggling playback performs a full teardown and recreation of the AudioUnit as follows:

- (IBAction)togglePlay:(UIButton *)selectedButton
    if (!toneUnit)
        // Create the audio unit as shown above
        [self createToneUnit];
        // Finalize parameters on the unit
        OSErr err = AudioUnitInitialize(toneUnit);
        NSAssert1(err == noErr, @"Error initializing unit: %ld", err);
        // Start playback
        err = AudioOutputUnitStart(toneUnit);
        NSAssert1(err == noErr, @"Error starting unit: %ld", err);
        [selectedButton setTitle:NSLocalizedString(@"Stop", nil) forState:0];
        // Tear it down in reverse
        toneUnit = nil;
        [selectedButton setTitle:NSLocalizedString(@"Play", nil) forState:0];


You can download the ToneGenerator and the complete sample project used in this post here (25kb)

The aim in this post was to present a sample iOS application that shows AudioUnits in the simplest way possible. Most of the Apple sample projects are considerably more elaborate, involve complex AUGraphs or can't be used on iOS because they use Mac-only APIs.

However, this sample application is a little atypical because of this simplification. In more complex applications, you'll probably want an AUGraph to chain an input node fed from a file, possibly a mixer and effects nodes, then an output node together.


Testing if an arbitrary pointer is a valid object pointer

In this post, I look at an approach for testing if an arbitrary pointer is a pointer to a valid Objective-C object. The result from the test is not absolutely accurate and can interfere with gdb debugging if the pointer isn't a valid memory location, so this is not something you'd want to do often (and certainly not in production code). But it can be a handy debugging tool for when you're staring blindly at memory you didn't allocate.


I originally wrote this code when I was looking through all the notifications from the local CFNotificationCenter, trying to work out where I was making mistakes in some AVFoundation video code (AVFoundation produces a lot of notifications when media playback is happening).

The callback function for these notifications has the following prototype:

void MyNotificationCallBack (
   CFNotificationCenterRef center,
   void *observer,
   CFStringRef name,
   const void *object,
   CFDictionaryRef userInfo

The object associated with the notification is passed as a void *. Most of the time, the value in object will be an Objective-C object but for the few times when it isn't, simply passing the result into NSLog can cause your program to crash.

What do you need to test?

Again, this comes back to the question: what is a valid Objective-C object?

For the purpose of sending a message to an object, all you need is for the value pointed to by the object pointer (i.e. the isa pointer) to be a registered Class value. This is the most important point to test — the chance of an arbitrary memory value being a valid Class by chance is fairly low (although certainly not zero).

But there is another requirement to be a usable object: the memory space following the isa pointer must be valid. This is not an easy thing to test, but it is possible to test if the object itself was allocated at least as large as the class' instance size.

However, this allocation test does not need to return positive for any object that wasn't heap allocated. The most common example of non-heap allocated objects in Objective-C are compiler created strings. So while appropriate heap allocation for a block of memory that starts with an isa pointer is a near guarantee that the object is a valid Objective-C object, failure of this test does not eliminate the possibility that it's a valid object.

Finally, while all of this is happening, there remains the possibility that the pointer you're testing doesn't point to a valid memory location at all. If you're interested in handling this situation, you need to set up a signal handler (or Mach exception handler) to catch SIGBUS and SIGSEGV signals (or EXC_BAD_ACCESS if you go the Mach exceptions route).

The code

#import <malloc/malloc.h>
#import <objc/runtime.h>

static sigjmp_buf sigjmp_env;

PointerReadFailedHandler(int signum)
    siglongjmp (sigjmp_env, 1);

BOOL IsPointerAnObject(const void *testPointer, BOOL *allocatedLargeEnough)
    *allocatedLargeEnough = NO;
    // Set up SIGSEGV and SIGBUS handlers
    struct sigaction new_segv_action, old_segv_action;
    struct sigaction new_bus_action, old_bus_action;
    new_segv_action.sa_handler = PointerReadFailedHandler;
    new_bus_action.sa_handler = PointerReadFailedHandler;
    new_segv_action.sa_flags = 0;
    new_bus_action.sa_flags = 0;
    sigaction (SIGSEGV, &new_segv_action, &old_segv_action);
    sigaction (SIGBUS, &new_bus_action, &old_bus_action);

    // The signal handler will return us to here if a signal is raised
    if (sigsetjmp(sigjmp_env, 1))
        sigaction (SIGSEGV, &old_segv_action, NULL);
        sigaction (SIGBUS, &old_bus_action, NULL);
        return NO;
    Class testPointerClass = *((Class *)testPointer);

    // Get the list of classes and look for testPointerClass
    BOOL isClass = NO;
    NSInteger numClasses = objc_getClassList(NULL, 0);
    Class *classesList = malloc(sizeof(Class) * numClasses);
    numClasses = objc_getClassList(classesList, numClasses);
    for (int i = 0; i < numClasses; i++)
        if (classesList[i] == testPointerClass)
            isClass = YES;

    // We're done with the signal handlers (install the previous ones)
    sigaction (SIGSEGV, &old_segv_action, NULL);
    sigaction (SIGBUS, &old_bus_action, NULL);
    // Pointer does not point to a valid isa pointer
    if (!isClass)
        return NO;
    // Check the allocation size
    size_t allocated_size = malloc_size(testPointer);
    size_t instance_size = class_getInstanceSize(testPointerClass);
    if (allocated_size > instance_size)
        *allocatedLargeEnough = YES;
    return YES;

Results from the function

Running this test program:

void LogPointerInformation(const void *somePointer)
    BOOL allocatedLargeEnough;
    BOOL isMessageableObject = IsPointerAnObject(somePointer, &allocatedLargeEnough);
    NSLog(@"The pointer %p is %@ and is %@.",
        isMessageableObject ?
            @"a valid object" :
            @"not a valid object",
        allocatedLargeEnough ?
            @"allocated at least as large as the required instance size" :
            @"not a known allocation");

int main (int argc, const char * argv[]) {
    NSAutoreleasePool * pool = [[NSAutoreleasePool alloc] init];

    LogPointerInformation([[[NSObject alloc] init] autorelease]);

    [pool drain];
    return 0;

Gives the following results (I've chopped off the NSLog times for brevity):

The pointer 0x100001130 is a valid object and is not a known allocation.
The pointer 0x10011e940 is a valid object and is allocated at least as large as the required instance size.
The pointer 0x100000c0d is not a valid object and is not a known allocation.
The pointer 0x12345678 is not a valid object and is not a known allocation.

Limitations of this approach

The most serious limitation of this approach is that it can never guarantee anything. For this reason, it is important that you never use this in production code.

The most obvious situation where the function will fail completely is when testing a malloc'd C array of Objective-C class pointers. This memory block definitely starts with a valid Class value and may even have a malloc_size greater than that class' instance size — but the block was never truly allocated as an object and if any instance values are important to the class, they are all likely to be invalid.

Objects that are not heap allocated are difficult to guarantee that their instance memory will be within addressable memory. This means that you could raise SIGSEGV or SIGBUS signals.

Still on the topic of signals, while I've included signal handling in this code, it's unlikely you'll ever want it to be invoked as it stops gdb dead.

While ordinarily, you can give gdb a handle signal nostop noprint pass to continue after the signal, that won't work here. There are issues getting the Mac version of gdb to proceed after an EXC_BAD_ACCESS. It is actually easier to let gdb catch the signal, then drag the program execution point in Xcode to the top of the sigsetjmp block manually.

As a final caveat: the signal handling code that I've written is strictly not thread-safe and non-reentrant.


The result is more of a heuristic than an absolute verdict.

However, the approach is a useful tool to give an "acceptable" verification that a value is a valid object for debugging purposes. It is certainly good enough to test before sending values to NSLog.

The issues with gdb when an invalid memory access signal is raised are annoying. I'd be interested to see if there's a way to avoid this. It certainly provides an added disincentive to using this code on truly arbitrary values.