An experimental plan to investigate message hiding in audio content
May 6, 2014
In this introductory post I describe my plan for performing a series of experiments centered on the Agilent 33622A to explore audio steganography and psychoacoustic masking. Over the spring and summer of 2014 I hope to carry out this experimental plan and report my findings here in the Test & Measurement blog space on Element 14. I want to get the plan posted now as I will be travelling in Europe for a while and won't be posting again until my return in June.
What are audio steganography and psychoacoutisc masking? They seem to me to be pretty much the same thing, though I may not have dug deep enough into the literature to discern any subtle differences. I will admit that both concepts, on first blush, appeared to be pseudoscientific nonsense to me, but I've since convinced myself that they are real phenomena and are worthy of experimental investigation. The basic idea in both is to exploit characteristics of human hearing so as to convey hidden information in audio content. So I will be trying to hide data in audio that can not be readily detected by human listeners. Now that may remind you of a bad spy thriller plot device, but my literature review indicates hiding intelligence in audio content actually works and psychoacoustic techniques are used extensively in commercial broadcasting to allow statistical tracking of listener habits. The ideas behind psychoacoustic masking are pretty simple, implementation, on the other hand, gets rather involved.
There is plenty of literature available on these topics. I recommend a Google search of the terms "audio steganography" and "psychoacoustic masking".
Because I'm a very curious fellow, I am compelled to investigate these ideas for myself. An instrument like the Agilent 33622A is well suited to this sort of experimenting because it can generate audio signals in the range of human hearing and it can sum in precisely separated and attenuated tones that form the backbone of psychoacoustic masking. The arbitrary generation capability with deep memory lends it self to experimentation with longer audio files and the modulation capabilities and 120 MHz frequency range allow experiments in low power AM and FM broadcasting.
My scheme is to proceed as follows:
- Use the 33622A and an appropriate speaker to verify that human hearing, at least my human hearing, can be fooled into not hearing things.
- This will involve experimenting with amplitude, frequency separation, phase relationships and duration of covert tone insertion to determine what does and does not get perceived.
- If part 1 is successful, see if I can mask a small amount of information into a short melodic audio file and detect the covert information using spectral analysis.
- Here the arbitrary waveform memory will come into play as storage for the short melodic file.
- If part 2 is successful, see if I can transmit the short melodic file using very low power AM and FM and detect the hidden information at the output of a receiver.
- Modulation capabilities on the 33622A combined with arbitrary waveform memory should allow experimental transmission of encoded signals.
Notes on experiments in psychoacoustic masking
These will be verification and characterization experiments that establish the realtionship between primary content and masked content. To keep things simple I will work with simple sinusoidal tones at first and try to establish the frequency proximity and amplitude relationships that allow covert tones to be psychoacoustically masked from listener perception. I have read articles and papers that describe findings in this area and I would like to verify the validity of reported findings.
Notes on experiments in steganographic content embedding
Again, staying with the simple philosophy I will work with one particlular source file composed of a series of sinusoidal tones in a simple melody. Working from the information I gather in the psychoacoustic masking experiements I will determine an ideal set of covert tones to insert into the source file then investgate two things. First, can a listener hear the covert tones? Second, can the covert content be detected on a spectrum analyzer?
Notes on experiments in very low power AM and FM broadcasting
Radio broadcast has fascinated me for a long time. I remember building a crystal radio in the early 1970's and being astonished that such a simple device could pick up and reproduce radio station broadcasts. During the summer of 1976 I worked for a local AM radio station as a night shift control room operator. One of the duties of the night shift control room operator was to switch the transmit power from 50 kW to 10 kW at dusk then back to 50 kW at dawn. As layers of the ionosphere cool at night skywave or skip propagation improves. A local broadcast may skip off the ionosphere at night and reach into distant markets where the signal may interfere with other licensed broadcasters. Reducing transmit power at night reduced the risk of interference. My point is that to be courteous I will be keeping the power very low in all broadcasting experiments.
The technology required to broadcast intelligible radio signals is fairly simple. For amplitude modulation (AM) broadcast all that is needed is a stable sinusoidal frequency source (the carrier) and circuitry that will allow an intelligence signal (music or voice for example), to modulate the amplitude of the carrier. Add a basic antenna and voila, you have an AM transmitter. Factors that affect the useful range of a transmitter include the power delivered to the antenna and the efficiency of the antenna design. Radio broadcasting is a tightly regulated industry with strict licensing requirements and serious oversight provided by federal agencies. Regulation and oversight are necessary to make sure all players can coexist and carry out their business in a fair and civilized marketplace. I emphasize these points because I want to caution readers to be very careful when building any electronics project that may purposefully or accidentally broadcast signals. You are responsible for the electromagnetic soup that spills out of your creations.
A waveform generator like the 33622A contains everything necessary to make a low power radio transmitter. It contains a highly stable sine wave generator to create a carrier that covers the range of AM and FM broadcast and it has a variety of modulation options suitable for impressing intelligence on the carrier signal. In this series of blog posts on audio steganography I plan to demonstrate very low power AM and FM broadcasting as a curiosity and illustrate why you should not indiscriminately play with radio broadcast technology. Frequencies from 535 kHz to 1605 kHz in the electromagnetic spectrum form the regulated AM commercial broadcast space in Canada. The frequency modulation (FM) band runs from 88 MHz to 108 MHz. Each band is portioned into channels with specified bandwidths, inter-channel spacing and transmit power limits. AM stations in Canada are spaced 10 kHz apart and FM stations are 200 kHz apart.
At any rate, what I hope to do in the blog post on transmitting within this series is demonstrate how the 33622A can be set up to broadcast a very low power AM or FM signal in an open channel over a distance of a few meters. I'd like to show how FM is less susceptible to amplitude disturbances that affect AM transmission. I would also like to illustrate the FM capture effect.
As with any research, it must be noted that the outcome might be a negative result at any stage. That is, at any stage in my plan I may not be able to show success. Realizing this, I will be open to changing the path of the investigation depending on what the findings indicate.
I hope readers will find my posts interesting and informative.
Mark
