Implementing Room acoustics correction using REW
Implementing Room acoustics correction using REW
While many people try improving the performance of their audio systems with higher-quality components, there is one component that most people miss: the listening room.
Normal listening rooms at home are rarely optimised for best sound reproduction. The best way would be to really optimise room acoustics. However, this often doesn’t work well as it would mean adding a lot of absorbers and diffusors to your listening room. The worst thing about it: To perform well in the low frequency range, these devices have to be very large.
But there’s another way to deal with problems in room acoustics – corrections using specifically designed electronic filters. This won’t be a full replacement for optimised room acoustics, but it will definitely improve the performance of you audio system.
What do you need?
- Raspberry Pi
- DSP equipped HiFiBerry board – the DAC+ DSP or the Beocreate 4 channel amplifier
- a measurement microphone
- a PC with an SPDIF output
- REW
- HiFiBerry DSP toolkit
While REW can run directly on the Raspberry Pi, we don’t recommend it. The Raspberry Pi is quite slow for this kinds of software which makes the usability quite bad.
Install the DSP toolkit on your Raspberry Pi
If the DSP toolkit isn’t already installed, you can install it with a simple script:
bash <(curl https://raw.githubusercontent.com/hifiberry/\ hifiberry-dsp/master/install-dsptoolkit)
Check if it is installed correctly by running “dsptoolkit”
dsptoolkit Can't read or parse /home/pi/.dsptoolkit/dspprogram.xml
Just ignore the error message at the moment, it’s important that the program itself is installed.
Install a DSP program with equalisation support
The DSP programs for B&O speakers and the “Basic fullrange” profile have been updated to support the equalisation feature.
If you have created your own DSP program and deployed it with SigmaStudio, you need to create a DSP profile first. Have a look at the documentation, how to create a DSP profile.
Before deploying filters, we strongly recommend to upload your DSP profile to the DSP first, even if it is already deployed. This is a safety measure. The reason for this: If the DSP profile deployed on the DSP is different from the profile that dsptoolkit is using, it might write settings to incorrect addresses on the DSP. In best case, nothing will happen, ins worst case you can mess up the DSP program completely. This could even destroy your speakers (unfortunately I know what I’m talking about).
The best way is to download the DSP profile XML file to the Raspberry Pi and install it using dsptoolkit:
dsptoolkit install-profile \ ./fullrange-extended.xml
The result should be somthing like
DSP profile ./fullrange-extended.xml installed Copied ./fullrange-extended.xml to ...
Now, you can be sure to work with the correct DSP profile.
Install REW
Download and install REW on your PC. While it also runs on the Raspberry Pi, we don’t recommend to install and run it on the RPi. The experience on the PC is much better as the program runs much smoother on a PC than on the Raspberry Pi.
Connect and position the microphone
If you use an USB measurement microphone (we recommend this), just connect it to an USB port of your PC. If you use an XLR measurement microphone, you will need an external microphone preamplifier to connect it to an input of your sound card. Don’t try to use cheap electret microphones with the onboard sound input from your PC. Most likely, linearity of both will be quite bad.
The microphone should be positioned vertically at your listening position. The height should be about the height of your head when you sit at this position. We recommend to use a microphone stand to position the microphone.
You might require an USB extension cable. Just note that for long extensions you will need an active USB extension cable. You can get these at your favourite computer shop.
Connect the DSP board
Connect the optical output of your PC or sound card to the optical input of the HiFiBerry DSP/DAC or 4-channel amplifier.
Configure REW
Make sure, you select your measurement microphone as the input device for REW in the setting dialogue. If your microphone comes with a calibration file, you can configure it under the “Mic/Meter” tab.
While you can configure the equalisation option later, we recommend also to set the correct equalisation options. You should select the Generic equalisation type:
Measure the frequency response on your listening position
Now create a new measurement by clicking the Measure button.
Make sure the frequency range is 20-20.000 Hz. During the measurement REW shows you the headroom of the mesurement. Ideally it should be between 3 and 10dB. If it’s more, increase the volume using the “Level (dbFS)” setting. If it is less than 3 or even 0dB, you need to reduce the output level.
When the measurement is finished, REW will show you the frequency response of your system:
Note that this the the frequency response at the position of your microphone. It will look differently if you move the microphone around. You can do multiple measurements at multiple positions and calculate an average to optimise the system for a broader range of listening positions.
In this measurement you can see that it is not linear. The biggest issue is the peak at 45Hz. This is looks like a so-called “room mode”. In general there are 2 reasons for a non-linear frequency response:
- Non-linearities of the speaker
- Interactions with the room
With a far-field measurement as we did it here, you can’t distinguish between both, but that’s not a problem. REW will correct both.
Calculate optimisation filters
Select the optimisation dialog by clicking on the EQ button.
Under “Equalizer” select “Generic”. Make sure speaker type is “Full range”.
Depending on your speakers, you can add a subsonic filter at a given frequency using the “Bass limited” setting. This is especially helpful with small speakers that can’t If you’re not sure what do do, leave it at 40Hz.
You might notice that the target isn’t perfectly linear. There is a slight increase of frequencies above 100Hz and a falling slope above 1kHz. The reason for this is again room acoustics. As the measurement took both direct and reflected sound into account, we need to reduce the high frequencies a bit. These frequencies are absorbed much more than lower frequencies. Therefore the measurement will measure less reflected sound here. A falling slope compensates for this effect.
First select the best matching “Target level” by clicking on “Set target level”. Then start the optimisation by clicking “Match response to target”. REW will now do it’s “magic” and calculate correction filters for your setup.
In the top-left diagram you will see the process, the correction filters and the expected frequency response.
Write the filters to the DSP
Now you only have to transfer the correction filters to your Raspberry Pi. First save these using the “Export filter settings as text” menu item. REW will now create a text file with the definition of the filters. It looks like this:
Filter Settings file Room EQ V5.18 Dated: Jul 5, 2018 12:03:08 PM Notes: Equaliser: Generic Jul 5 12:01:46 Filter 1: ON PK Fc 7515 Hz Gain -9.7 dB Q 1.00 Filter 2: ON PK Fc 1911 Hz Gain -3.1 dB Q 1.00 Filter 3: ON PK Fc 50.4 Hz Gain -6.0 dB Q 2.58 Filter 4: ON PK Fc 58.9 Hz Gain 2.5 dB Q 2.00 Filter 5: ON PK Fc 43.4 Hz Gain -3.8 dB Q 3.74 Filter 6: ON PK Fc 352 Hz Gain -2.9 dB Q 2.98 Filter 7: ON PK Fc 156 Hz Gain -4.2 dB Q 4.49 Filter 8: ON PK Fc 7193 Hz Gain 9.0 dB Q 2.09 Filter 9: ON PK Fc 3388 Hz Gain 3.8 dB Q 2.47 Filter 10: ON PK Fc 13666 Hz Gain -3.2 dB Q 4.98 Filter 11: ON PK Fc 706 Hz Gain -2.8 dB Q 1.50 Filter 12: ON PK Fc 675 Hz Gain 3.8 dB Q 5.00 Filter 13: ON PK Fc 590 Hz Gain 3.0 dB Q 5.00 Filter 14: ON None Filter 15: ON None Filter 16: ON None Filter 17: ON None Filter 18: ON None Filter 19: ON None Filter 20: ON None
Transfer this file to your Raspberry Pi (I recommend scp for this), e.g.
scp filters.txt pi@ip-of-your-pi
Login to your Raspberry Pi and write these filters to the DSP using the DSP Toolkit:
dsptoolkit apply-rew-filters filter.txt
Run another measurement
If you run another measurement now, the frequency response should match the target response quite well. It won’t be perfectly flat, but that’s fine.
Save the settings on the Raspberry Pi
You just downloaded the settings to the DSP, but these will be lost after a reboot. Therefore, you should save the DSP settings on the Raspberry Pi.
sudo -E dsptoolkit store-filters
Note that the “-E” command line option is important here. Otherwise dsptoolkit might not find the configuration files.
The result should look like this:
Settings stored to /etc/dspparameter.dat DSP program copied to /etc/dspprogram.xml
This step is important to make sure the correct filters will be loaded when the system powers up as the changes of filters won’t be stored in the EEPROM of the DSP.
Remove the filters
Not happy with the results and you want to start with a new measurement and another optimisation? You can remove all filter settings using the DSP toolkit
sudo dsptoolkit clear-iir-filters
Tips & tricks
- Run the optimization with a signal on both speakers.
If your speakers are identical, you should apply the same filters to both speakers, especially if you use the full frequency range (20-20.000Hz). - Use equalisation only where it is needed.
Most problems will be in the low frequencies (0-200Hz). If you’re speakers frequency response is reasonably flat (it doesn’t need to be perfect!), optimise only the low frequencies. Slight peaks of 2-3dB in the frequency response are not a problem. You will see these with practically every speaker in a normal listening environment. Don’t over-optimise!
Low-frequency vs. full range optimisations
The following pictures shows 2 equalisation setting: One full-range optimisation and one only in the range of 20-200Hz.
Which one sound better? In both cases you will still hear the “personality” of your speakers, but they will sound a bit different. Try different profiles and listen to them. There is no simple right or wrong – at the end your personal preference is important.