In this business we often talk about how easy it is to get computers to talk to each other; computers without networks are almost inconceivable. Despite being standardized as little as five years ago, we now expect them all to play nicely together.

Even in the home, a CAT-5 connection isn't too remarkable, but mobile devices still spend most of their time in lonely isolation.

Of course, many technologies exist for connecting handheld devices to networks and other devices, but costs and limitations often cause people to question if it's really necessary: Why should a handheld be talking to the world? Well, I came up with one idea, and this article shows how I implemented it.

It starts with the conversion of my CD collection to MP3 to make it easier to find tracks and automate playback.

Once your collection is on your computer, you need software to organize it and play it back; many freely available applications will do that with varying amounts of control.

Random playback is all very well, but it can lead to some very strange combinations if your collection is broad (jumping from Beethoven to Aerosmith can be something of a shock), and it occurred to me that it should be possible to come up with something better. As the owner of a PDA (actually, several, but for the moment it's a Compaq iPaq) with a Wi-Fi card, I thought it would be nice to control my playback wirelessly.

My requirements were pretty basic; random play does work for me most of the time, but I would like to be able to pause playback and adjust the volume. There are also occasions when I'd like to select the next track to be played - when the mood takes me or when talking to guests about music - so some facility for browsing through the available tracks to select one would be good.

I should also point out that I'm very lazy, and the idea of entering the details of each recorded track doesn't appeal to me; the application must be able to cope with additional tracks being added automatically and notice if any have been removed.

Given this set of requirements, it's obviously important that all my music is stored in a standard format, something that most of the MP3 ripping software is ideally suited to. I had used an application called Audio Catalyst, but this currently has some problems with Windows 2000, so now I use AudioGrabber, which works well for my needs. This application grabs tracks from a CD, converts (rips) them to MP3 format, picks up the track listing information over the Internet, and is able to provide the MP3 file names that correspond to the track names. Because of the way I intend to use the tracks, I decided to organize them into directories by bands, then into subdirectories by album, and finally form the track name from the name of the band and the name of the track separated with a hyphen ("-"). Thus a sample listing looks something like this:

Blur\Park Life\Blur - Badhead.mp3
Blur\Park Life\Blur - Bank Holiday.mp3
Blur\Park Life\Blur - Clover Over Dover.mp3
... etc. ...

Once I decided on how the music was to be stored and what my requirements were for playing it back, I started working out what my application would look like and how it would work. A formal design process probably would have helped at this point, but since the application was only intended for my use (something that will be obvious if you care to examine the source code, which can be downloaded from below), I felt that as long as I had the overall structure defined, I could get on with coding, something I came to regret later, as you will see.

Obviously, the application would have two major components: a server and a client. The server would have a minimum GUI, just enough to ensure it was working properly, while the client would have controls for changing the volume, pausing playback, and selecting the next track (see Figure 1).

The inclusion of a GUI on the server side was, strictly speaking, unnecessary, as it should be easy to run up a second client on the server if local control is needed. I also decided that it was important for the server and the client not to be in constant contact for two reasons: first, the battery life on an iPaq using a Wi-Fi link is laughable, measured in minutes not hours (in Part 2 or 3 of this article I'll discuss moving the application over to Bluetooth with its much improved power consumption), so it was important to be able to turn the client off without affecting playback. Second, I wanted to be able to run several clients with the same server.

The motivation for this second requirement is less clear. I have several computers and a wife with her own PDA, so while the thought of being in sole control of our listening choices appealed to me, it would be less popular with others in the house.

This decision has several implications: primarily, it was not possible for the client to constantly display the track playing currently. I decided the easiest solution was to add a button to update the display on request (see Figure 2). It was possible to set up some sort of registration system for clients and establish an outgoing connection from the server to all interested clients whenever a new track started, but I decided this was too complex at the moment; it could always be added later if it seemed important.

It was clear that objects would represent the tracks, and on the server side these objects would exist in some sort of ArrayList (I like ArrayLists).

The intention was that on launch, once fed with a directory to start from, the server would scan this directory and any subdirectories looking for MP3 files that it could use to build the ArrayList. Requests to play a specific track were then indexed by its position in the ArrayList. Clearly, this meant that tracks could not be added without stopping and relaunching the server, but this was a minor point and not considered important.

The first stage was to consider the networking model: What kind of communication would take place and what kind of responses would be expected? As all communication was client-initiated, it was obvious that the server would be using a ServerSocket listening on a specific port (1710, for no other reason than it happens to be my birthday). The client would connect to that port and send a request, the server would respond, and the client would break the connection. While it's possible to send several requests and receive several responses in one go, it was not deemed necessary (this is not a very time-critical application).

On connection the server should respond with a message to confirm that it exists ("+OK"), then wait for the client to send a request for it to respond to. I adopted the Internet convention of starting all successful responding messages with a "+", while those that gave an error of any type start with a "-"; this allows the client to simply check the first character of a response to see if the request has been honored. I also decided that all network communications would be case-insensitive. While this can slow things down slightly, it makes testing easier and avoids irritating problems with the use of the wrong case. I set about mapping out all the network communication I'd be using, starting with the ability to display which tracks were being played on the client:

[server] +OK
[client] status
[server] +Last Playing 292#Clannad - Ta Me Mo Shui
[server] +Playing 866#Men Without Hats
- 01 - In the 21st Century
[server] +Next 1057#Steeleye Span - The Victory

The response is split with the "#" symbol, the number before it indicates the position of the track in the ArrayList held on the server. This is used to identify the track when requests are sent back to the server from the client. It was possible to use the track name, and then search the ArrayList for a matching track (and, indeed, this proved necessary in another context), but it seemed easier at the time to use numbers. There was no attempt to optimize the protocol to reduce network traffic. Given the short requests and responses this seemed unnecessary, and a human-readable protocol makes for easier testing.

The other commands are displayed in much the same way, allowing the user to pause, resume, adjust the volume, and play the next and last tracks. I did decide that the response to the status request would be useful in many other contexts, so several of the commands respond with the same list.

A request to play the previous track would lead to a lag as this track would not be prepared to play unless it maintained its prepared state, something I'd have to decide later on once I knew how long the lag was and if it was tolerable (as it turned out it's only a couple of seconds, which I can tolerate).

This covered the basic commands, but I also wanted to be able to browse the available tracks and pick one to be played, so I had to add commands to retrieve a track listing and to play a specific track indexed by number.

The track listing returns the file names of the tracks held on the server in the order in which they occur in the ArrayList (so no index number is needed). It also returns the full filename of the track. The client, on receipt of this data, can copy it into a local vector (no ArrayList in PersonalJava) with the knowledge that the index numbers will match those used on the server. The data is terminated with a full-stop (".") alone on a line, the same as the SMTP protocol.

The play request is numerically indexed; the word "play" is followed by a space and the number of the track to be played. This track then replaces the previously considered next track, and the server responds with a status response listing the last, current, and next tracks. If the intention is to start the chosen track immediately, a "next" message will need to be sent.

At this stage it's important to document your network protocol in as much detail as possible. I went to the extent of printing up examples and pinning them to the wall above my desk. Any changes to the protocol should be recorded immediately and extensions added to the written record.

When you're working with a team of programmers this is common sense, but when you're working alone it's easy to think you'll remember all the details, and then promptly come unstuck later when you discover you can't.

Having worked out what the client and the server are going to do, it was time to start writing them. Getting the JMF installed and working was very smooth, and while the API can appear confusing, it's actually easy to get the hang of it. By making the tracks responsible for their own preparation and playback, I was able to encapsulate all the JMF functions into a single class. Probably the most complex part is preparing the track for playing, which is accomplished with the following code:

details = new File(fname);
URL u = new URL("file:///" + fname);
Manager.setHint(Manager.PLUGIN_PLAYER, new Boolean(true));
p = Manager.createPlayer(u);
p.realize();
p.prefetch();
p.addControllerListener(this);

The String variable fname is simply the full path and filename of the MP3 file. The final line indicates that the Track class itself is the one to be informed of playback events (such as when the track finishes or has completed its preparation). Once the track is prepared, the following methods allow control from the main application:

public void start() {
p.start();
}
public void pause() {
p.stop();
}
public void stop() {
p.stop();
p.setMediaTime(new Time(0));
}

It's also important to be able to release the resources being used by a track for playback, which was done with the following code:

public void release() {
loaded = false;
p.stop();
p.deallocate();
p.close();
}

This way a track could exist in the ArrayList using few resources, then when it was known that it would be played soon, it would be prepared to start when the previous track had completed. The Boolean loaded is used to check if a track has been loaded. This is set to true on receipt of a PrefetchCompleteEvent that's passed to the ControllerListener (as specified during track preparation).

I also decided that Track objects should be responsible for deciding if they would play, so I added another method called willYouPlay(), which returns the Boolean "true" every time. This was intended to provide a more complex choice mechanism that I added later and will discuss in Part 2 of this article.

Once I got the server working, I tested it using Hyperterminal (connecting to 127.0.0.1, the TCP/IP loopback address), then set about creating a client. When developing applications such as this it's always best to start with the server, so it can be tested using a terminal application.

On the client side I developed the application using normal Java, being careful to use only constructs that are available in PersonalJava (so no beloved ArrayList).

While PersonalJava is a cutdown version of Java, as long as you're not using Swing and check before using any of the more recent APIs, you can develop in Java and then test in PersonalJava without too many problems. In this respect I did pretty well, falling down only on remembering to use elementAt when accessing a vector. The basic controls were easy enough with a simple GUI and buttons; the only issue was with picking up button click events using Personal- Java on the iPaq (the Sun implementation, though the problem also manifested itself using JEODE). Clicks don't always make themselves heard, and I was eventually forced to switch to MouseDown events to ensure usable interaction.

This had the side effect of making the buttons less responsive (they don't depress when clicked), but functionality won out over appearance.

Getting the track browsing and selection to work was considerably more effort, just working out what the interface should look like was a task in itself. I started thinking that a treeview would work well, but the thought of having to write one held me back, and I wasn't convinced that it would scale well (by this time I was working with several hundred CDs comprising over 1,600 tracks). I eventually went for two lists, both held in a single dialog frame (remember that in PersonalJava you can have only one dialog and one frame visible at a time, so the main application is in a frame and the PickTracks object is a dialog). Album names are held in the top list, while track names are displayed at the bottom (updated when a different album is selected) (see Figure 3). I decided against having any buttons in this part of the GUI as simply clicking (or tapping, as we will be using a pen) will be sufficient to select the next track to be played.

This gave me a working application. I used the PersonalJava emulation environment to make sure it was all compatible (and thus discovered the vector problem, which was easily fixed) and transferred the application to my iPaq. It works; I can now listen to music and control what comes next, but that is only the beginning...

In Part 2 of this article I'll show how I added a profile to each track, enabling me to set my mood and have suitable tracks chosen by preference. I'll also show how Bluetooth is a better network solution for this kind of application, and how it can be made to work with Java. In Part 3 I'll discuss how I added the next stage of control and enabled my mobile device to switch speakers on and off around my house.

Useful Links